A multimodal framework for question answering

Abstract

Systems and techniques are described herein for generating answers to questions. For instance, a method for generating answers to questions is provided. The method may include detecting an object in a scene of a user; obtaining scene data associated with the scene; generating scene information based on the detected object and the scene data; identifying a region of interest (ROI) of an image of the scene; generating user information based on a question and user data associated with the user; processing the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identifying a video from among a plurality of videos that is relevant to the answer; and providing at least one of the answer or the video to the user.

Description

Qualcomm Ref. No. 2502633 WO1A MULTIMODAL FRAMEWORK FOR QUESTION ANSWERING CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of Indian Provisional Application No.202521017860, filed February 28, 2025, which is incorporated herein by reference in its entirety.TECHNICAL FIELD

[0002] The present disclosure generally relates to answering questions. For example, aspects of the present disclosure include systems and techniques for generating answers to questions.BACKGROUND

[0003] Recent advancements in artificial intelligence (Al) and machine-learning (ML) technologies have led to the development of increasingly sophisticated models capable of understanding and interpreting complex data structures. One class of such models is referred to large generative Al models or large generative machine-learning models (LXMs). LXMs have a multitude of applications that span across various domains, from natural language processing to computer vision and speech recognition. Their efficacy stems from their abi 1 i ty to learn from massive datasets, gaining an unprecedented depth of understanding and applicability.

[0004] The increasing capabilities of LXMs, including (but not limited to) Large Language Models (LLMs), Large Speech Models (LSMs), and Large Vision Models (LVMs) (which are also referred to as Language Vision Models or Vision Language Models (VLMs)), offer enhanced functionality in various applications such as natural language understanding, speech recognition, visual analysis, text generation, speech generation, image generation, and / or the like. Among the diverse t pes of LXMs, LLMs are generally known for their capabilities in understanding and generating human language. These models may be trained on extensive textual datasets and may perform such tasks as machine translation, text summarization, question-answering, and / or the like. LLMs have found applications in a broad range of industries including healthcare, finance, and customer service, among others.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO2

[0005] An LVM is a LXM that is trained to interpret and analyze visual data. LVM models may use convolutional neural networks or similar architectures to process visual inputs and derive meaningful conclusions from them. From image classification to object detection and generating new images in response to natural language prompts, LVMs are growing in populanty and use in diverse areas such as medical imaging, autonomous vehicles, surveillance systems, advertising, and entertainment.

[0006] A large multimodal model (LMM) may be trained to receive different types of inputs and generate responses based on the inputs. For example, a LMM may process image data and text data to generate an output. For instance, an LMM may respond to a text query’ about an image (for example, the query may be “how many people are in this image?’’ or “identify the people in this image”). As another example, an LMM may respond to an instruction relative to a video (such as “summarize this video”).

[0007] Extended reality (XR) technologies can be used to present virtual content to users, and / or can combine real environments from the physical world and virtual environments to provide users with XR experiences. The term XR can encompass virtual reality (VR), augmented reality (AR), mixed reality (MR), and the like. XR systems can allow users to experience XR environments by overlaying virtual content onto a user’s view of a real -world environment. For example, an XR head-mounted device (HMD) may include a display that allows a user to view the user’s real-world environment through a display of the HMD (e.g., a transparent display). The XR HMD may display virtual content at the display in the user’s field of view overlaying the user’s view of their real-world environment. Such an implementation may be referred to as “see-through” XR. As another example, an XR HMD may include a scene-facing camera that may capture images of the user’s real-world environment. The XR HMD may modify or augment the images (e.g., adding virtual content) and display the modified images to the user. Such an implementation may be referred to as “pass through” XR or as “video see through (VST).” The user can generally change their view of the environment interactively, for example by tilting or moving the XR HMD.SUMMARY

[0008] The following presents a simplified summary relating to one or more aspects disclosed herein. Thus, the following summary should not be considered an extensivePolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO3overview relating to all contemplated aspects, nor should the following summan be considered to identify key or critical elements relating to all contemplated aspects or to delineate the scope associated with any particular aspect. Accordingly, the following summary presents certain concepts relating to one or more aspects relating to the mechanisms disclosed herein in a simplified form to precede the detailed description presented below.

[0009] Systems and techniques are described for generating answers to questions. According to at least one example, a method is provided for generating answers to questions. The method includes: detecting an object in a scene of a user; obtaining scene data associated with the scene; generating scene information based on the detected object and the scene data; identifying a region of interest (ROI) of an image of the scene; generating user information based on a question and user data associated with the user; processing the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identifying a video from among a plurality of videos that is relevant to the answer; and providing at least one of the answer or the video to the user.

[0010] In another example, an apparatus for generating answers to questions is provided that includes at least one memory and at least one processor (e.g., configured in circuitry) coupled to the at least one memory. The at least one processor configured to: detect an object in a scene of a user; obtain scene data associated with the scene; generate scene information based on the detected object and the scene data; identify a region of interest (ROI) of an image of the scene; generate user information based on a question and user data associated with the user; process the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identify a video from among a plurality of videos that is relevant to the answer; and provide at least one of the answer or the video to the user.

[0011] In another example, a non-transitory computer-readable medium is provided that has stored thereon instructions that, when executed by one or more processors, cause the one or more processors to: detect an object in a scene of a user; obtain scene data associated with the scene; generate scene information based on the detected object and the scene data; identify a region of interest (ROI) of an image of the scene; generate user information based on a question and user data associated with the user; process the Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO4question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identify a video from among a plurality7of videos that is relevant to the answer; and provide at least one of the answer or the video to the user.

[0012] In another example, an apparatus for generating answers to questions is provided. The apparatus includes: means for detecting an object in a scene of a user; means for obtaining scene data associated with the scene; means for generating scene information based on the detected object and the scene data; means for identifying a region of interest (ROI) of an image of the scene; means for generating user information based on a question and user data associated with the user; means for processing the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; means for identifying a video from among a plurality7of videos that is relevant to the answer; and means for providing at least one of the answ er or the video to the user.

[0013] In some aspects, one or more of the apparatuses described herein is, can be part of, or can include an extended reality device (e.g., a virtual reality (VR) device, an augmented reality (AR) device, or a mixed reality (MR) device), a vehicle (or a computing device, system, or component of a vehicle), a mobile device (e.g., a mobile telephone or so-called “smart phone”, a tablet computer, or other ty pe of mobile device), a smart or connected device (e.g., an Intemet-of-Things (loT) device), awearable device, a personal computer, a laptop computer, a video server, a television (e.g., a network-connected television), a robotics device or system, or other device. In some aspects, each apparatus can include an image sensor (e.g., a camera) or multiple image sensors (e.g., multiple cameras) for capturing one or more images. In some aspects, each apparatus can include one or more displays for displaying one or more images, notifications, and / or other displayable data. In some aspects, each apparatus can include one or more speakers, one or more light-emitting devices, and / or one or more microphones. In some aspects, each apparatus can include one or more sensors. In some cases, the one or more sensors can be used for determining a location of the apparatuses, a state of the apparatuses (e.g., a tracking state, an operating state, a temperature, a humidity level, and / or other state), and / or for other purposes.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO5

[0014] This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification of this patent, any or all drawings, and each claim.

[0015] The foregoing, together with other features and aspects, will become more apparent upon referring to the following specification, claims, and accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Illustrative examples of the present application are described in detail below with reference to the following figures:

[0017] FIG. 1 is a diagram illustrating an example extended-reality (XR) system, according to aspects of the disclosure;

[0018] FIG. 2 is a diagram illustrating another example extended reality (XR) system, according to aspects of the disclosure;

[0019] FIG. 3 is a diagram illustrating yet another example extended-reality (XR) system, according to aspects of the disclosure;

[0020] FIG. 4 is a block diagram illustrating an architecture of an example extended reality (XR) system, in accordance with some aspects of the disclosure;

[0021] FIG. 5 is a block diagram illustrating an example system for generating answers to questions, according to various aspects of the present disclosure;

[0022] FIG. 6 is a block diagram illustrating an example system for generating answers to questions, according to various aspects of the present disclosure;

[0023] FIG. 7 is a block diagram of an example system that may generate output based on inputs, according to various aspects of the present disclosure;

[0024] FIG. 8 is a block diagram illustrating an example system for generating an answer to respond to a question, according to various aspects of the present disclosure;

[0025] FIG. 9 is a block diagram illustrating an example system for generating an answer to respond to a question, according to various aspects of the present disclosure;Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO6

[0026] FIG. 10 is a flow diagram illustrating an example process that an answer selector may use to select an answer from among a plurality of answers, according to various aspects of the present disclosure;

[0027] FIG. 11 is a flow diagram illustrating an example process for generating answers to questions, in accordance with aspects of the present disclosure;

[0028] FIG. 12 is a block diagram illustrating an example of a deep learning neural network that can be used to perform various tasks, according to some aspects of the disclosed technology;

[0029] FIG. 13 is a block diagram illustrating an example of a convolutional neural network (CNN), according to various aspects of the present disclosure;

[0030] FIG. 14 includes an example machine-learning model that may be used in various aspects of the present disclosure;

[0031] FIG. 15 is a block diagram illustrating a multimodal generative ML system for generating natural language responses based on natural language input from a prompt and any additional information;

[0032] FIG. 16 is a block diagram of an example transformer in accordance with some aspects of the disclosure; and

[0033] FIG. 17 is a block diagram illustrating an example computing-device architecture of an example computing device which can implement the various techniques described herein.DETAILED DESCRIPTION

[0034] Certain aspects of this disclosure are provided below. Some of these aspects may be applied independently and some of them may be applied in combination as would be apparent to those of skill in the art. In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of aspects of the application. However, it will be apparent that various aspects may be practiced without these specific details. The figures and description are not intended to be restrictive.

[0035] The ensuing description provides example aspects only, and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO7description of the exemplary aspects will provide those skilled in the art with an enabling description for implementing an exemplary aspect. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the application as set forth in the appended claims.

[0036] The terms “exemplary"’ and / or “example” are used herein to mean “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” and / or "‘example” is not necessarily to be construed as preferred or advantageous over other aspects. Likewise, the term “aspects of the disclosure” does not require that all aspects of the disclosure include the discussed feature, advantage, or mode of operation.

[0037] As noted previously, an extended reality (XR) system or device can provide a user with an XR experience by presenting virtual content to the user (e.g., for a completely immersive experience) and / or can combine a view of a real-world or physical environment with a display of a virtual environment (made up of virtual content). The real-world environment can include real-world objects (also referred to as physical objects), such as people, vehicles, buildings, tables, chairs, and / or other real-world or physical objects. As used herein, the terms XR system and XR device are used interchangeably. Examples of XR systems or devices include head-mounted displays (HMDs) (which may also be referred to as a head-mounted devices), XR glasses (e g., AR glasses, MR glasses, etc.) (also referred to as smart or network-connected glasses), among others. In some cases, XR glasses are an example of an HMD. In some cases, an XR system can track parts of the user (e.g.. a hand and / or fingertips of a user) to allow the user to interact with items of virtual content.

[0038] XR systems can include virtual reality (VR) systems facilitating interactions with VR environments, augmented reality (AR) systems facilitating interactions with AR environments, mixed reality' (MR) systems facilitating interactions with MR environments, and / or other XR systems.

[0039] For instance, VR provides a complete immersive experience in a three-dimensional (3D) computer-generated VR environment or video depicting a virtual version of a real-world environment. VR content can include VR video in some cases, which can be captured and rendered at very high quality, potentially providing a truly immersive virtual reality' experience. Virtual reality' applications can include gaming, training, education, sports video, online shopping, among others. VR content can be Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO8rendered and displayed using a VR system or device, such as a VR HMD or other VR headset, which fully covers a user’s eyes during a VR experience.

[0040] AR is a technology that provides virtual or computer-generated content (referred to as AR content) over the user’s view of a physical, real-world scene or environment. AR content can include virtual content, such as video, images, graphic content, location data (e.g., global positioning system (GPS) data or other location data), sounds, any combination thereof, and / or other augmented content. An AR system or device is designed to enhance (or augment), rather than to replace, a person’s current perception of reality. For example, a user can see a real stationary' or moving physical object through an AR device display, but the user's visual perception of the physical object may be augmented or enhanced by a virtual image of that object (e.g., a real-world car replaced by a virtual image of a DeLorean), by AR content added to the physical object (e.g., virtual wings added to a live animal), by AR content displayed relative to the physical object (e.g., informational virtual content displayed near a sign on a building, a virtual coffee cup virtually anchored to (e.g., placed on top of) a real-world table in one or more images, etc.), and / or by displaying other types of AR content. Various types of AR systems can be used for gaming, entertainment, and / or other applications.

[0041] MR technologies can combine aspects of VR and AR to provide an immersive experience for a user. For example, in an MR environment, real-world and computergenerated objects can interact (e.g., a real person can interact with a virtual person as if the virtual person were a real person).

[0042] An XR environment can be interacted with in a seemingly real or physical way. As a user experiencing an XR environment (e.g., an immersive VR environment) moves in the real world, rendered virtual content (e.g., images rendered in a virtual environment in a VR experience) also changes, giving the user the perception that the user is moving within the XR environment. For example, a user can turn left or right, look up or down, and / or move forwards or backwards, thus changing the user’s point of view of the XR environment. The XR content presented to the user can change accordingly, so that the user’s experience in the XR environment is as seamless as it would be in the real world.

[0043] In some cases, an XR system can match the relative pose and movement of objects and devices in the physical world. For example, an XR system can use tracking information to calculate the relative pose of devices, objects, and / or features of the real- Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO9world environment in order to match the relative position and movement of the devices, objects, and / or the real-world environment. In some examples, the XR system can use the pose and movement of one or more devices, objects, and / or the real-world environment to render content relative to the real-world environment in a convincing manner. The relative pose information can be used to match virtual content with the user’s perceived motion and the spatio-temporal state of the devices, objects, and real-world environment. In some cases, an XR system can track parts of the user (e.g., a hand and / or fingertips of a user) to allow the user to interact with items of virtual content.

[0044] XR systems or devices can facilitate interaction with different types of XR environments (e.g., a user can use an XR system or device to interact with an XR environment). One example of an XR environment is a metaverse virtual environment. A user may virtually interact with other users (e.g., in a social setting, in a virtual meeting, etc.), virtually shop for items (e.g., goods, services, property; etc.), to play computer games, and / or to experience other services in a metaverse virtual environment. In one illustrative example, an XR system may provide a 3D collaborative virtual environment for a group of users. The users may interact with one another via virtual representations of the users in the virtual environment. The users may visually, audibly, haptically, or otherwise experience the virtual environment while interacting with virtual representations of the other users.

[0045] A virtual representation of a user may be used to represent the user in a virtual environment. A virtual representation of a user is also referred to herein as an avatar. An avatar representing a user may mimic an appearance, movement, mannerisms, and / or other features of the user. In some examples, the user may desire that the avatar representing the person in the virtual environment appear as a digital twin of the user. In any virtual environment, it is important for an XR system to efficiently generate high-quality avatars (e.g., realistically representing the appearance, movement, etc. of the person) in a low-latency manner. It can also be important for the XR system to render audio in an effective manner to enhance the XR experience.

[0046] In some cases, an XR system can include an optical “see-through” or “pass-through” display (e.g., see-through or pass-through AR HMD or AR glasses), allowing the XR system to display XR content (e.g., AR content) directly onto a real-world view without displaying video content. For example, a user may view physical objects through Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO10a display (e.g., glasses or lenses), and the AR system can display AR content onto the display to provide the user with an enhanced visual perception of one or more real -world objects. In one example, a display of an optical see-through AR system can include a lens or glass in front of each eye (or a single lens or glass over both eyes). The see-through display can allow the user to see a real-world or physical object directly, and can display (e.g., projected or otherwise displayed) an enhanced image of that object or additional AR content to augment the user’s visual perception of the real world.

[0047] The term “recurrent neural network” (RNN) is used herein to refer to a class of neural networks particularly well-suited for sequence data processing. Unlike feedforward neural networks, RNNs may include cycles or loops within the network that allow information to persist. This enables RNNs to maintain a “memory” of previous inputs in the sequence, which may be beneficial for tasks in which temporal dynamics and the context in which data appears are relevant.

[0048] The term “long short-term memory network” (LSTM) is used herein to refer to a specific type of RNN that addresses some of the limitations of basic RNNs, particularly the vanishing gradient problem. LSTMs include a more complex recurrent unit that allows for the easier flow of gradients during backpropagation. This facilitates the model’s ability to learn from long sequences and remember over extended periods, making it apt for tasks such as language modeling, machine translation, and other sequence-to-sequence tasks.

[0049] The term “transformer” is used herein to refer to a specific type of neural network that includes an encoder and / or a decoder and can be well-suited for sequence data processing. Transformers may use multiple self-attention components to process input data in parallel rather than sequentially. The self-attention components may be configured to weigh different parts of an input sequence when producing an output sequence. Unlike solutions that focus on the relationship between elements in two different sequences, self-attention components may operate on a single input sequence. The self-attention components may compute a weighted sum of all positions in the input sequence for each position, which may allow the model to consider other parts of the sequence when encoding each element. This may offer advantages in tasks that benefit from understanding the contextual relationships between elements in a sequence, such as sentence completion, translation, and summarization. The weights may be learned during the training phase, allowing the model to focus on the most contextually relevant parts of Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO11the input for the task at hand. Transformers, with their specialized architecture for handling sequence data and their capacity for parallel computation, often serve as foundational elements in constructing large generative Al models (LXM).

[0050] The term “large generative Al model” (LXM) is used herein to refer to an advanced computational framework that includes any of a variety of specialized Al models including, but not limited to, large language models (LLMs), large speech models (LSMs), large / language vision models (LVMs), vision language models (VLMs), hybrid models, and multi-modal models (which may be referred to as large multimodal models (LMMs)). An LXM may include multiple layers of neural networks (e.g., recurrent neural network (RNN), long short-term memory network. (LSTM), transformer, etc.) with millions or billions of parameters. Unlike traditional systems that translate user prompts into a series of correlated files or web pages for navigation, LXMs support dialogic interactions and encapsulate expansive knowledge in an internal structure. As a result, rather than merely serving a list of relevant websites, LXMs are capable of providing direct answers and / or are otherwise adept at various tasks, such as text summarization, translation, complex question-answering, conversational agents, etc. In various aspects, LXMs may operate independently as standalone units, may be integrated into more comprehensive systems and / or into other computational units (e.g., those found in a SoC or SIP, etc.), and / or may interface with specialized hardware accelerators to improve performance metrics such as latency and throughput. In some aspects, the LXM component may be enhanced with or configured to perform an adaptive algorithm that allows the LXM to better understand context information and dynamic user behavior. In some aspects, the adaptive algorithms may be performed by the same processing system that manages the core functionality of the LXM and / or may be distributed across multiple independent processing systems.

[0051] LXMs are generally trained with a large corpus of data (e.g., public data). When LXMs are asked to answer questions (e.g., at an inference phase of operation) the LXMs may not have specific details of a current context to give answers relevant to the current context. Further, the training data may be outdated and incorrect over time

[0052] RAG models address at least some of these issues. RAG models may retrieve the most relevant recent data pertaining to a question. Further RAG models may augmentPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO12inputs to an LXM, then used the LXM to generate answers that are relevant to a current context.

[0053] RAG models may obtain personalized info and store the personalized data in a document. The RAG models may divide the document into chunks of roughly similar size. Further the RAG models may encode each chunk into a dense feature vector using encoders like miniature language model (MiniLM) or (BAII) General Embedding (BGE). The RAG models may store the feature vector in a vector database (e.g., a RAG server).

[0054] RAG models may further encode a question (e.g., in a text format) from a user using the same encoder and select the top-K chunks with maximum feature similarity to the question text. The RAG model may retrieve the relevant chunks using variations of Maximum Inner Product Score (MIPS) to establish the similarity (e.g., a “relevance score" or “similarity score’7). The similarity is returned back to the application. The RAG model may then augment the question text with the retrieved chunks and input to the LLM for answer generation.

[0055] Users have personal preferences and constraints (e g., dietary restrictions, allergies, etc.). These preferences and constraints can affect daily choices regarding food, drink, movies, clothes, books etc.

[0056] To obtain goods and services, users often visit different vendors (e.g., in person or online). Goods and / or services that are in line with a user's preferences and constraints may not be available at even- place / website visit by the user and / or when the user desires the goods and / or sen ices.

[0057] In order to obtain goods and / or services that are in line with a user's personal preferences and / or constraints, the user may obtain a menu or browse a website of a vendor, evaluate options available from the vendor, read through details of the options (e.g., ingredients and / or materials), read through pricing information regarding the options, find the best match with the user's preferences and constraints, and finally order the selected goods and / or services.

[0058] When buying some items like equipment, it will also be useful to see a demonstrative video. Such videos may be retrieved from a customer database or from a manufacturer or vendor of the equipment.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO13

[0059] Altogether, it may take a lot of time to acquire goods and sendees that are in line with a user's preferences and / or constraints. For example, there may be significant time between when a user enters a shop (or visits a website) to when the user orders an item.

[0060] Systems, apparatuses, methods (also referred to as processes), and computer-readable media (collectively referred to herein as “systems and techniques”) are described herein for generate answers to questions. For example, the systems and techniques described herein may generate an answer to a question based on user data (e.g., based on a person that asked the question) and based on scene data (e.g. based on an environment of the user when they asked the question).

[0061] For example, the systems and techniques may include or implement a personalized assistant for an XR device that has information regarding a user's needs and vendor's available items and recommends the appropriate items (e.g., available items that are in line with the user's preferences and constraints) to the user. The systems and techniques may also provide appropriate videos to guide the customer (e.g, instructional, demonstrative, and / or review videos).

[0062] The systems and techniques may involve the use of object detection as a preprocessing for a scene retrieval augmented generation (RAG). The scene RAG may be related to the scene. For example, the scene RAG may be related to a retail location or website. The scene RAG may obtain information regarding the scene from an own er or operator of the scene (e.g., the retail location or website). For example, the scene RAG may retrieve relevant sections of the vendor offerings for a spatio-temporal context.

[0063] The systems and techniques may use open-vocabulary detection (OVD) for capturing spatial context and use the spatial context for a scene RAG along with a User Profile. The scene RAG may be adapted to the context. The systems and techniques use open-vocabulary-based object detection which generalizes better to in the wild objects.

[0064] Additionally or alternatively, the systems and techniques may map LXM answers to videos from a database. For example, the systems and techniques may use a RAG and Out of Context detection to identify video through relevance scores.

[0065] In some aspects, the systems and techniques may include a double or triple RAG framework including a user RAG, a scene RAG and a video RAG. The user RAG may be Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO14used to generate user information based on stored user preferences and / or constraints. The scene RAG may be used to generate scene information based on goods or sendees available from an own er or operator of the scene. The video RAG may be used to identify videos based on a similarity between an answer to a query of a user and titles or descriptors of available videos.

[0066] Additionally or alternatively, the systems and techniques may process data using the scene RAG and the user RAG in parallel. Processing data using the scene RAG and the user RAG in parallel may reduce the latency for answer generation from an LXM.

[0067] In some aspects, the systems and techniques may select one or more regions of interest (ROIs). Further the systems and techniques may provide the ROIs to provide context for an LXM to provide accurate answers to questions posed by a user (e.g., in a RAG framework).

[0068] The systems and techniques may use a single resolution to enhance visual features of image data by getting relevant ROIs from an open- vocabulary object detection framework which may use fewer tokens than other systems.

[0069] In some aspects, the systems and techniques may implement a negationdetection technique. The negation-detection technique may operate on questions posed by a user and answers provided by an LXM through low-cost string-matching (e.g., in a RAG framework).

[0070] In some aspects, the systems and techniques may implement a string matching on both Questions and Answers in a RAG system to answer questions (e.g., to recommend products / services to users). The systems and techniques take aflfixal negations into account explicitly for English language.

[0071] The systems and techniques may be used in, for example, shopping (e.g., providing recommendations either online or in store), providing medical and healthcare recommendations, personalized navigation, personalized recommendations in agentic Al, prompting relevant actions in a spatio-temporal context. The systems and techniques may be useful in the fields of, as examples, retail, education, and / or healthcare.

[0072] Various aspects of the application will be described with respect to the figures below.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO15

[0073] FIG. 1 is a diagram illustrating an example extended-real ity (XR) system 100, according to aspects of the disclosure. As shown. XR system 100 includes an XR device 102. XR device 102 may implement, as examples, image-capture, object-detection, object-tracking, gaze-tracking, view-tracking, localization (e.g., determining a location of XR device 102), pose-tracking (e.g., tracking a pose of XR device 102 and / or a pose of one or more objects in scene 112), content-generation, content-rendering, computational, communicational, and / or display aspects of extended reality, including virtual reality7(VR). augmented reality (AR), and / or mixed reality (MR).

[0074] For example, XR device 102 may include one or more scene-facing cameras that may capture images of a scene 112 in which a user 108 uses XR device 102. XR device 102 may detect and / or track objects (e.g., object 114) in scene 112 based on the images of scene 112. In some aspects, XR device 102 may include one or more userfacing cameras that may capture images of eyes of user 108. XR device 102 may determine a gaze of user 108 based on the images of user 108. In some aspects, XR device 102 may determine an object of interest (e.g., object 114) in scene 112 (e.g., based on the gaze of user 108, based on object recognition, and / or based on a received indication regarding object 114). XR device 102 may obtain and / or render XR content 116 (e.g., text, images, and / or video) for display at XR device 102. XR device 102 may display XR content 116 to user 108 (e.g., within a field of view 110 of user 108). In some aspects, XR content 116 may be based on and / or anchored to points in scene 112. For example, XR content 116 may be, or may include, an altered version of object 114 (e.g., based on an XR application running at XR device 102) anchored to object 114 in scene 112. The XR application may provide user 108 with an XR experience by altering scene 112 in view 110 of user 108. In some aspects, XR device 102 may display XR content 116 in relation to the view of user 108 of the object of interest. For example, XR device 102 may overlay XR content 116 onto object 114 in field of view 110. In any case, XR device 102 may overlay XR content 116 (whether related to object 114 or not) onto the view of user 108 of scene 112. For example, object 114 may be a cherry tree. Based on an XR application running at XR device 102, XR device 102 may anchor XR content 116, which may be a palm tree, to object 114 such that in the view of user 108, user 108 sees XR content 116 (the palm tree) and not object 114 (the cherry tree).Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO16

[0075] In a “see-through” or “transparent” configuration, XR device 102 may include a transparent surface (e.g., optical glass) such that XR content 116 may be displayed on (e.g., by being projected onto) the transparent surface to overlay the view of user 108 of scene 112 as viewed through the transparent surface. In a “pass-through” configuration or a “video see-through” configuration, XR device 102 may include a scene-facing camera that may capture images of scene 112. XR device 102 may display images or video of scene 112, as captured by the scene-facing camera, and XR content 116 overlaid on the images or video of scene 112.

[0076] In various examples, XR device 102 may be, or may include, a head-mounted device (HMD), a virtual reality headset, and / or smart glasses. XR device 102 may include one or more cameras, including scene-facing cameras and / or user-facing cameras, a GPU, one or more sensors (e.g., such as one or more inertial measurement units (IMUs), image sensors, and / or microphones), one or more communication units (e.g., wireless communication units), and / or one or more output devices (e.g., such as speakers, headphones, display, and / or smart glass).

[0077] In some aspects, XR device 102 may be, or may include, two or more devices. For example, XR device 102 may include a display device and a processing device. The display device may capture and / or generate data, such as image data (e.g., from userfacing cameras and / or scene-facing cameras) and / or motion data (from an inertial measurement unit (IMU)). The display device may provide the data to the processing device, for example, through a wireless connection between the display device and the processing device. The processing device may process the data and / or other data (e.g., data received from another source). Further, the processing unit may generate (or obtain) XR content 116 to be displayed at the display device. The processing device may provide the generated XR content 116 to the display device, for example, through the wireless connection. And the display device may display XR content 116 in field of view7110 of user 108.

[0078] FIG. 2 is a diagram illustrating an example extended reality (XR) system 200, according to aspects of the disclosure. In some aspects, an XR system may be, or may include, two or more devices. The two or more devices of XR system 200 may perform the operations described with regard to XR system 100 of FIG. 1.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO17

[0079] For example, XR system 200 includes a display device 204 and a processing device 206. In some aspects, display device 204 and processing device 206 may implement a communication link 210 between display device 204 and processing device 206. Communication link 210 may be a wireless connection according to any suitable wireless protocol, such as, a broadband-cellular-network protocol, for example, a fifth generation (5G) wireless cellular protocol.

[0080] In some aspects, XR system 200 may include a companion device 208. Display device 204 and companion device 208 and may implement a communication link 212 between display device 204 and companion device 208 and companion device 208 and processing device 206 may implement a communication link 214 between companion device 208 and processing device 206. Communication link 212 may be a wireless connection according to any suitable wireless protocol, such as, for example, Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.15, or Bluetooth®. Communication link 214 may be a wireless connection according to any suitable wireless protocol, such as, a broadband-cellular-network protocol, for example, a fifth generation (5G) wireless cellular protocol.

[0081] Display device 204, processing device 206, and / or companion device 208 may collectively implement as examples, image-capture, object-detection, object-tracking, gaze-tracking, view-tracking, localization, pose-tracking, content-generation, contentrendering, computational, communicational. and / or display aspects ofXR. For example, display device 204 may implement image-capture, gaze-tracking, view-tracking, localization, pose-tracking, communicational, and / or display aspects of XR. Processing device 206 may implement object-detection, object-tracking, localization, contentgeneration, content-rendering, computational, and / or communicational, aspects of XR. Additionally or alternatively, companion device 208 may implement at least a portion of one or more of localization, pose-tracking, communicational, object-detection, objecttracking, localization, content-generation, content-rendering, and / or computational aspects of XR.

[0082] For example, display device 204 may capture and / or generate data, such as image data (e.g., from user-facing cameras and / or scene-facing cameras) and / or motion data (from an inertial measurement unit (IMU)). Display device 204 may provide the dataPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO18to processing device 206, for example, through communication link 210 or through communication link 212, companion device 208, and communication link 214.

[0083] Processing device 206 may process the data and / or other data (e.g., data received from another source or data stored at processing device 206). For example, processing device 206 may detect, recognize, and / or track objects in scene 218 based on the images of scene 218. Further, processing device 206 may generate (or obtain) XR content 220 to be rendered for display at display device 204. Processing device 206 may render XR content 220 to be appropriate for display at display device 204 (e.g., based on a pose of display device 204). Processing device 206 may provide rendered XR content 220 to display device 204 through communication link 210 (or communication link 214, companion device 208, and communication link 212) and display device 204 may display XR content 220 in field of view 216 of user 202.

[0084] In various examples, display device 204 may be, or may include, a headmounted display (HMD), a virtual reality headset, and / or smart glasses. Display device 204 may include one or more cameras, including scene-facing cameras and / or user-facing cameras, a GPU. one or more sensors (e.g., such as one or more inertial measurement units (IMUs), image sensors, and / or microphones), and / or one or more output devices (e.g., such as speakers, headphones, displays, and / or smart glass). In other examples, display device 204 may include a handheld device with a display, such as a smartphone or tablet.

[0085] Processing device 206 may be, or may include, for example, a server computer (e.g., an edge or cloud-based server, a personal computer acting as a server device, or a mobile device acting as a server device). Processing device 206 may be configured to store virtual content and / or perform operations related to rendering the virtual content as image data suitable for providing to display device 204 for display. Companion device 208 may be, or may include, a smartphone, laptop, tablet computer, personal computer, gaming system, any other computing device and / or a combination thereof.

[0086] FIG. 3 is a diagram illustrating an example extended-reality (XR) system 300, according to aspects of the disclosure. As shown, XR system 300 includes an XR device 302 including a display 304. In some cases, XR device 302 may implement, as examples, image-capture, object detection, gaze-tracking, view-tracking, computational and / or display aspects of extended reality, including virtual reality (VR), augmented reality Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO19(AR), and / or mixed reality (MR). For example, XR device 302 may include one or more scene-facing cameras that may capture images of a scene in which a user 308 uses XR device 302. XR device 302 may detect objects in the scene based on the images of the scene. Further, XR device 302 may include one or more user-facing cameras that may capture images of eyes of user 308. XR device 302 may determine a gaze of user 308 based on the images of user 308. XR device 302 may determine an object of interest in the scene based on the gaze of user 308. XR device 302 may obtain and / or render information (e.g., text, images, and / or video based on the object of interest). XR device 302 may display the information to a user 308 at display 304 (e.g., within a field of view 310 of user 308).

[0087] XR device 302 may operate in in a “pass-through’’ configuration. For example, XR device 302 may include a scene-facing camera that may capture images of the scene of user 308. XR device 302 may display images or video of the scene, as captured by the scene-facing camera, and information overlaid on the images or video of the scene. XR device 302 may display the information to be viewed by a user 308 in field of view 310 of user 308. For example, in a “see-through” configuration, XR device 302 may include a transparent surface (e.g., optical glass) such that information may be displayed on the transparent surface to overlay the information onto the scene as viewed through the transparent surface.

[0088] XR device 302 and / or display 304 may be, or may include, a handheld device, a smartphone, a tablet, or another computing device with a display. XR device 302 include one or more cameras, including scene-facing cameras and / or user-facing cameras, a GPU, one or more sensors (e.g., such as one or more inertial measurement units (IMUs), image sensors, and / or microphones), and / or one or more output devices (e.g., such as speakers, display, and / or smart glass).

[0089] FIG. 4 is a diagram illustrating an architecture of an example extended reality (XR) system 400, in accordance with some aspects of the disclosure. XR system 400 may execute XR applications and implement XR operations.

[0090] In this illustrative example, XR system 400 includes one or more image sensors 402, an accelerometer 404, a gyroscope 406, storage 408, an input device 410, a display 412, Compute components 414, an XR engine 426, an image processing engine 428, a rendering engine 430, and a communications engine 432. It should be noted that the Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO20components 402-432 shown in FIG. 4 are non-limiting examples provided for illustrative and explanation purposes, and other examples may include more, fewer, or different components than those shown in FIG. 4. For example, in some cases, XR system 400 may include one or more other sensors (e.g., one or more inertial measurement units (IMUs), radars, light detection and ranging (LIDAR) sensors, radio detection and ranging (RADAR) sensors, sound detection and ranging (SODAR) sensors, sound navigation and ranging (SONAR) sensors, audio sensors, etc.), one or more display devices, one more other processing engines, one or more other hardware components, and / or one or more other software and / or hardware components that are not shown in FIG. 4. While various components of XR system 400, such as image sensor 402, may be referenced in the singular form herein, it should be understood that XR system 400 may include multiple of any component discussed herein (e.g., multiple image sensors 402).

[0091] Display 412 may be, or may include, aglass, ascreen, alens, aprojector, and / or other display mechanism that allows a user to see the real-world environment and also allows XR content to be overlaid, overlapped, blended with, or otherwise displayed thereon.

[0092] XR system 400 may include, or may be in communication with, (wired or wirelessly) an input device 410. Input device 410 may include any suitable input device, such as a touchscreen, a pen or other pointer device, a keyboard, a mouse a button or key, a microphone for receiving voice commands, a gesture input device for receiving gesture commands, a video game controller, a steering wheel, a joystick, a set of buttons, a trackball, a remote control, any other input device discussed herein, or any combination thereof. In some cases, image sensor 402 may capture images that may be processed for interpreting gesture commands.

[0093] XR system 400 may also communicate with one or more other electronic devices (wired or wirelessly). For example, communications engine 432 may be configured to manage connections and communicate with one or more electronic devices. In some cases, communications engine 432 may correspond to communication interface 1726 of FIG. 17.

[0094] In some implementations, image sensors 402, accelerometer 404, gyroscope 406, storage 408, display 412, compute components 414, XR engine 426, image processing engine 428, and rendering engine 430 may be part of the same computing Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO21device. For example, in some cases, image sensors 402, accelerometer 404, gyroscope 406, storage 408, display 412, compute components 414, XR engine 426, image processing engine 428, and rendering engine 430 may be integrated into an HMD, extended reality glasses, smartphone, laptop, tablet computer, gaming system, and / or any other computing device. However, in some implementations, image sensors 402, accelerometer 404, gyroscope 406, storage 408, display 412, compute components 414, XR engine 426, image processing engine 428, and rendering engine 430 may be part of two or more separate computing devices. For instance, in some cases, some of the components 402-432 may be part of, or implemented by. one computing device and the remaining components may be part of, or implemented by, one or more other computing devices. For example, such as in a split perception XR system, XR system 400 may include a first device (e.g., an HMD), including display 412, image sensor 402, accelerometer 404. gyroscope 406, and / or one or more compute components 414. XR system 400 may also include a second device including additional compute components 414 (e.g., implementing XR engine 426, image processing engine 428, rendering engine 430, and / or communications engine 432). In such an example, the second device may generate virtual content based on information or data (e.g., images, sensor data such as measurements from accelerometer 404 and gyroscope 406) and may provide the virtual content to the first device for display at the first device. The second device may be, or may include, a smartphone, laptop, tablet computer, personal computer, gaming system, a sen- er computer or ser er device (e.g., an edge or cloud-based server, a personal computer acting as a server device, or a mobile device acting as a server device), any other computing device and / or a combination thereof.

[0095] Storage 408 may be any storage device(s) for storing data. Moreover, storage 408 may store data from any of the components of XR system 400. For example, storage 408 may store data from image sensor 402 (e.g., image or video data), data from accelerometer 404 (e.g., measurements), data from gyroscope 406 (e.g., measurements), data from compute components 414 (e.g., processing parameters, preferences, virtual content, rendering content, scene maps, tracking and localization data, object detection data, privacy data, XR application data, face recognition data, occlusion data, etc.), data from XR engine 426, data from image processing engine 428, and / or data from renderingPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO22engine 430 (e.g., output frames). In some examples, storage 408 may include a buffer for storing frames for processing by compute components 414.

[0096] Compute components 414 may be, or may include, a central processing unit (CPU) 416, a graphics processing unit (GPU) 418, a digital signal processor (DSP) 420, an image signal processor (ISP) 422, a neural processing unit (NPU) 424, which may implement one or more trained neural networks, and / or other processors. Compute components 414 may perform various operations such as image enhancement, computer vision, graphics rendering, extended reality operations (e.g., tracking, localization, pose estimation, mapping, content anchoring, content rendering, predicting, etc ), image and / or video processing, sensor processing, recognition (e.g., text recognition, facial recognition, object recognition, feature recognition, tracking or pattern recognition, scene recognition, occlusion detection, etc.), trained machine-learning operations, filtering, and / or any of the various operations described herein. In some examples, compute components 414 may implement (e.g., control, operate, etc.) XR engine 426, image processing engine 428, and rendering engine 430. In other examples, compute components 414 may also implement one or more other processing engines.

[0097] Image sensor 402 may include any image and / or video sensors or capturing devices. In some examples, image sensor 402 may be part of a multiple-camera assembly, such as a dual-camera assembly. Image sensor 402 may capture image and / or video content (e.g., raw image and / or video data), which may then be processed by compute components 414. XR engine 426, image processing engine 428, and / or rendering engine 430 as described herein.

[0098] In some examples, image sensor 402 may capture image data and may generate images (also referred to as frames) based on the image data and / or may provide the image data or frames to XR engine 426, image processing engine 428, and / or rendering engine 430 for processing. An image or frame may include a video frame of a video sequence or a still image. An image or frame may include a pixel array representing a scene. For example, an image may be a red-green-blue (RGB) image having red, green, and blue color components per pixel; a luma, chroma-red, chroma-blue (Y CbCr) image having a luma component and two chroma (color) components (chroma-red and chroma-blue) per pixel; or any other suitable type of color or monochrome image.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO23

[0099] In some cases, image sensor 402 (and / or other camera of XR system 400) may be configured to also capture depth information. For example, in some implementations, image sensor 402 (and / or other camera) may include an RGB-depth (RGB-D) camera. In some cases, XR system 400 may include one or more depth sensors (not shown) that are separate from image sensor 402 (and / or other camera) and that may capture depth information. For instance, such a depth sensor may obtain depth information independently from image sensor 402. In some examples, a depth sensor may be physically installed in the same general location or position as image sensor 402 but may operate at a different frequency or frame rate from image sensor 402. In some examples, a depth sensor may take the form of a light source that may proj ect a structured or textured light pattern, which may include one or more narrow bands of light, onto one or more objects in a scene. Depth information may then be obtained by exploiting geometrical distortions of the projected pattern caused by the surface shape of the object. In one example, depth information may be obtained from stereo sensors such as a combination of an infra-red structured light projector and an infra-red camera registered to a camera (e.g., an RGB camera).

[0100] XR system 400 may also include other sensors in its one or more sensors. The one or more sensors may include one or more accelerometers (e.g., accelerometer 404), one or more gyroscopes (e.g.. gyroscope 406). and / or other sensors. The one or more sensors may provide velocity, orientation, and / or other position-related information to compute components 414. For example, accelerometer 404 may detect acceleration by XR system 400 and may generate acceleration measurements based on the detected acceleration. In some cases, accelerometer 404 may provide one or more translational vectors (e.g.. up / down, left / right. forward / back) that may be used for determining a position or pose of XR system 400. Gyroscope 406 may detect and measure the orientation and angular velocity of XR system 400. For example, gyroscope 406 may be used to measure the pitch, roll, and yaw of XR system 400. In some cases, gyroscope 406 may provide one or more rotational vectors (e.g., pitch, yaw, roll). In some examples, image sensor 402 and / or XR engine 426 may use measurements obtained by accelerometer 404 (e.g., one or more translational vectors) and / or gyroscope 406 (e.g., one or more rotational vectors) to calculate the pose of XR system 400. As previously- noted, in other examples, XR system 400 may also include other sensors, such as aPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO24magnetometer, a gaze and / or eye tracking sensor, a machine vision sensor, a smart scene sensor, a speech recognition sensor, an impact sensor, a shock sensor, a position sensor, a tilt sensor, etc.

[0101] As noted above, in some cases, the one or more sensors may include at least one IMU. An IMU is an electronic device that measures the specific force, angular rate, and / or the orientation of XR system 400. using a combination of one or more accelerometers, one or more gyroscopes, and / or one or more magnetometers. For example, an IMU of XR system 400 may include accelerometer 404, gyroscope 406, and / or a magnetometer. In some examples, the one or more sensors may output measured information associated with the capture of an image captured by image sensor 402 (and / or other camera of XR system 400) and / or depth information obtained using one or more depth sensors of XR system 400.

[0102] The output of one or more sensors (e.g., accelerometer 404, gyroscope 406, and / or other sensors) can be used by XR engine 426 to determine a pose of XR system 400 (also referred to as the head pose) and / or the pose of image sensor 402 (or other camera of XR system 400). In some cases, the pose of XR system 400 and the pose of image sensor 402 (or other camera) can be the same. The pose of image sensor 402 refers to the position and orientation of image sensor 402 relative to a frame of reference (e g., with respect to a field of view 110 of FIG. 1). In some implementations, the camera pose can be determined for 6-Degrees of Freedom (6DoF). which refers to three translational components (e.g., which can be given by X (horizontal). Y (vertical), and Z (depth) coordinates relative to a frame of reference, such as the image plane) and three angular components (e.g. roll, pitch, and yaw relative to the same frame of reference). In some implementations, the camera pose can be determined for 3-Degrees of Freedom (3DoF), which refers to the three angular components (e.g. roll, pitch, and yaw).

[0103] In some cases, a device tracker (not shown) can use the measurements from the one or more sensors and image data from image sensor 402 to track a pose (e.g., a 6DoF pose) of XR system 400. For example, the device tracker can fuse visual data (e.g., using a visual tracking solution) from the image data with inertial data from the measurements to determine a position and motion of XR system 400 relative to the physical world (e.g., the scene) and a map of the physical world. As described below, in some examples, when tracking the pose of XR system 400, the device tracker can generate a three-dimensional Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO25(3D) map of the scene (e.g., the real world) and / or generate updates for a 3D map of the scene. The 3D map updates can include, for example and without limitation, new or updated features and / or feature or landmark points associated with the scene and / or the 3D map of the scene, localization updates identifying or updating a position of XR system 400 within the scene and the 3D map of the scene, etc. The 3D map can provide a digital representation of a scene in the real / physical world. In some examples, the 3D map can anchor position-based objects and / or content to real-world coordinates and / or objects. XR system 400 can use a mapped scene (e.g., a scene in the physical world represented by, and / or associated with, a 3D map) to merge the physical and virtual worlds and / or merge virtual content or objects with the physical environment.

[0104] In some aspects, the pose of image sensor 402 and / or XR system 400 as a whole can be determined and / or tracked by compute components 414 using a visual tracking solution based on images captured by image sensor 402 (and / or other camera of XR system 400). For instance, in some examples, compute components 414 can perform tracking using computer vision-based tracking, model -based tracking, and / or simultaneous localization and mapping (SLAM) techniques. For instance, compute components 414 can perform SLAM or can be in communication (wired or wireless) with a SLAM system (not shown). SLAM refers to a class of techniques where a map of an environment (e.g., a map of an environment being modeled by XR system 400) is created while simultaneously tracking the pose of a camera (e.g., image sensor 402) and / or XR system 400 relative to that map. The map can be referred to as a SLAM map which can be three-dimensional (3D). The SLAM techniques can be performed using color or grayscale image data captured by image sensor 402 (and / or other camera of XR system 400) and can be used to generate estimates of 6DoF pose measurements of image sensor 402 and / or XR system 400. Such a SLAM technique configured to perform 6DoF tracking can be referred to as 6DoF SLAM. In some cases, the output of the one or more sensors (e.g., accelerometer 404, gyroscope 406, and / or other sensors) can be used to estimate, correct, and / or otherwise adjust the estimated pose.

[0105] In some cases, the 6DoF SLAM (e.g., 6DoF tracking) can associate features observed from certain input images from the image sensor 402 (and / or other camera) to the SLAM map. For example, 6DoF SLAM can use feature point associations from an input image to determine the pose (position and orientation) of the image sensor 402Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO26and / or XR system 400 for the input image. 6D0F mapping can also be performed to update the SLAM map. In some cases, the SLAM map maintained using the 6D0F SLAM can contain 3D feature points triangulated from two or more images. For example, key¬ frames can be selected from input images or a video stream to represent an observed scene. For every key frame, a respective 6D0F camera pose associated with the image can be determined. The pose of the image sensor 402 and / or the XR system 400 can be determined by projecting features from the 3D SLAM map into an image or video frame and updating the camera pose from verified 2D-3D correspondences.

[0106] In one illustrative example, the compute components 414 can extract feature points from certain input images (e.g., every input image, a subset of the input images, etc.) or from each key frame. A feature point (also referred to as a registration point) as used herein is a distinctive or identifiable part of an image, such as a part of a hand, an edge of a table, among others. Features extracted from a captured image can represent distinct feature points along three-dimensional space (e.g., coordinates on X, Y, and Z-axes), and every feature point can have an associated feature location. The feature points in key frames either match (are the same or correspond to) or fail to match the feature points of previously-captured input images or key frames. Feature detection can be used to detect the feature points. Feature detection can include an image processing operation used to examine one or more pixels of an image to determine whether a feature exists at a particular pixel. Feature detection can be used to process an entire captured image or certain portions of an image. For each image or key frame, once features have been detected, a local image patch around the feature can be extracted. Features may be extracted using any suitable technique, such as Scale Invariant Feature Transform (SIFT) (which localizes features and generates their descriptions), Learned Invariant Feature Transform (LIFT), Speed Up Robust Features (SURF), Gradient Location-Orientation histogram (GLOH), Oriented Fast and Rotated Brief (ORB), Binary Robust Invariant Scalable Keypoints (BRISK), Fast Retina Keypoint (FREAK), KAZE, Accelerated KAZE (AKAZE), Normalized Cross Correlation (NCC), descriptor matching, another suitable technique, or a combination thereof.

[0107] As one illustrative example, the compute components 414 can extract feature points corresponding to a mobile device, or the like. In some cases, feature points corresponding to the mobile device can be tracked to determine a pose of the mobilePolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO27device. As described in more detail below, the pose of the mobile device can be used to determine a location for projection of AR media content that can enhance media content displayed on a display of the mobile device.

[0108] In some cases, the XR system 400 can also track the hand and / or fingers of the user to allow the user to interact with and / or control virtual content in a virtual environment. For example, the XR system 400 can track a pose and / or movement of the hand and / or fingertips of the user to identify or translate user interactions with the virtual environment. The user interactions can include, for example and without limitation, moving an item of virtual content, resizing the item of virtual content, selecting an input interface element in a virtual user interface (e.g., a virtual representation of a mobile phone, a virtual keyboard, and / or other virtual interface), providing an input through a virtual user interface, etc.

[0109] FIG. 5 is a block diagram illustrating an example system 500 for generating answers to questions, according to various aspects of the present disclosure. For example, a user 502 may pose a question and XR device 504 may generate and present an answer to the question. According to various aspects of the present disclosure, XR device 504 may generate the answer based on a scene 506 in which user 502 uses XR device 504 and / or based on object(s) 508 in scene 506. XR device 504 may be an example of XR device 102 of FIG. 1, display device 204 of FIG. 2 or XR device 302 of FIG. 3.

[0110] FIG. 6 is a block diagram illustrating an example system 600 for generating answers to questions, according to various aspects of the present disclosure. For example, a user (e.g., user 502) may pose a question and system 600 may generate and present an answer to the question. According to various aspects of the present disclosure, system 600 may generate the answer based on a scene (e.g., scene 506) in which the user uses system 600 and / or based on objects (e.g., object(s) 508) in the scene.

[0111] Image sensor 602 may be a scene-facing camera of a device (e.g., XR device 504 of FIG. 5, of XR device 102 of FIG. 1, display device 204 of FIG. 2 or XR device 302 of FIG. 3). Image sensor 602 may capture image data 604 representative of the scene (e.g., scene 506) in which the device is used.

[0112] Object detector 606 may detect objects (e.g., object(s) 508) in the scene based on image data 604 and generate detected objects 608 indicative of the detected objects.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO28Object detector 606 may be, or may include, a machine-learning model trained to detect, identify, classify, and / or track objects in images, object detector 606 may be, or may include, an open-vocabulary detector (OVD) and / or an open-vocabulary tracker (OVT). Detected objects 608 may be, or may include, labels of detected objects.

[0113] Region of interest (ROI) determiner 610 may determine one or more ROIs 612 of image data 604. ROI determiner 610 may determine ROIs 612 based on a gaze of the user (e.g., based on a gaze tracker determining a gaze of a user based on images of the eyes of the user captured by eye-facing cameras of the device), a center of image data 604 (e.g., a center of a field of view (FoV) of image sensor 602), text detected in the image, and / or objects detected in the image (e.g., as indicated by detected objects 608).

[0114] Scene data storage 614 may be a data storage associated with the scene (e.g., scene 506). Scene data storage 614 may be remote from the scene. Scene data storage 614 may be, or may include, a database hosted by a remote server. The device may access (e.g., through a wireless connection, for example, across a network such as the Internet) scene data storage 614 to obtain scene data 616. Scene data 616 may be, or may include, data related to the scene. For example, if the scene is a retail venue, scene data 616 may include inventory data, price data, specifications, ingredients, etc. related to products and / or sen ices associated with the retail venue.

[0115] In some aspects. System 600 may identify scene data storage 614, for example, System 600 may identify' a data storage associated with the scene in which image sensor 602 captures image data 604. For example, in some aspects, System 600 (e.g., object detector 606) may detect a quick-response (QR) code, a name associated with the scene, and / or a recognizable landmark in the image of the scene. System 600 may further identify' the scene based on the quick-response (QR) code, the name associated with the scene, and / or the recognizable landmark. System 600 may identify scene data storage 614 as a data storage associated with the scene based on identification of the scene.

[0116] Scene data 616 may be obtained from, for example, a menu, product offerings, an inventory status, staffing constraints, etc. Scene data 616 may be obtained by, for example, querying a vendor database with all info already chunked and embedded. For example, when a user (e.g., user 502) enters a shop, user 502 may scans a QR code in the shop. Based on scanning the QR code, the device may request scene data 616 from scene data storage 614 and scene data storage 614 may provide scene data 616.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO29

[0117] Scene RAG 618 may process detected objects 608 and scene data 616 to generate scene information 620. Scene RAG 618 may be an example of RAG system 706 of FIG. 7. For example, scene RAG 618 may include an encoder and may use the encoder to encode detected objects 608 to generate features. Similarly, scene RAG 618 may use the encoder to encode scene data 616 to generate features. Scene RAG 618 may compare the features based on detected objects 608 with the features based on scene data 616 to determine a similarity between detected objects 608 and scene data 616.

[0118] Scene information 620 may be, or may include, information based on detected objects 608 and scene data 616. For example, scene information 620 may include information relevant to objects in the FoV of image sensor 602. For instance, scene information 620 may include price data, specifications, and / or ingredients of objects that are available and in an FoV of image sensor 602.

[0119] Microphone 622 may be, or may include, one or more microphones of the device. Microphone 622 may capture utterances of a user of the device and translate the utterances into audio data 624.

[0120] Text generator 626 may be, or may include, a speech-to-text converter. For example, text generator 626 may be, or may include, an LSM. Text generator 626 may generate question 628 based on audio data 624.

[0121] User data storage 630 may be. or may include, a data storage associated with the user (e.g., user 502). For example, user data storage 630 may store user data 632, which may include a profile, preferences, and / or constraints of the user. In some aspects, user data storage 630 may be included in the device (e.g., in a memory of XR device 504). In other aspects, the device may remotely access user data storage 630. In any case, the device may access user data storage 630 to obtain user data 632.

[0122] User data 632 may be obtained from, for example, medical records, diary entries, purchase receipts, shopping history, etc. of the user. User data 632 may be obtained by, for example, document scans by taking photos (e.g., using image sensor 602) or from online-browsing record.

[0123] User RAG 634 may process question 628 and user data 632 to generate user information 636. user RAG 634 may be an example of RAG system 706 of FIG. 7 For example, user RAG 634 may include an encoder and may use the encoder to encode Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO30question 628 to generate features. Similarly, user RAG 634 may use the encoder to encode user data 632 to generate features, user RAG 634 may compare the features based on question 628 with the features based on user data 632 to determine a similarity between question 628 and user data 632. User RAG 634 may chunk and embed user data 632.

[0124] User information 636 may be, or may include, information based on question 628 and user data 632. For example, user information 636 may include information of user data 632 that is relevant to question 628. For instance, user information 636 may include contextual information that may be useful to answering question 628.

[0125] LXM 638 may be, or may include, a large multimodal model (LMM). LXM 638 may be trained to generate answers to questions. In some aspects, LXM 638 may be trained to generate the answers to the questions based on contextual information (e.g., provided as prompts with the question). In some aspects, LXM 638 may be trained to generate answers based on multimodal inputs (e.g., text inputs, image inputs, video inputs, audio inputs, etc.).

[0126] In some aspects, system 600 may format image data 604, ROI 612, scene information 620, user information 636 as contextual information in a prompt. The prompt may further include question 628. For example, the prompt may instruct LXM 638 to generate answer 640 to question 628 based on contextual information represented in image data 604, ROI 612, scene information 620, and / or user information 636. In some aspects, LXM 638 may be trained with specific inputs assigned for image data 604, ROI 612, scene information 620, and / or user information 636.

[0127] In any case, LXM 638 may generate answer 640 based on image data 604, ROI 612, scene information 620, question 628, and user information 636. Answer 640 may be an answer to question 628 based on image data 604. ROI 612, scene information 620. and / or user data 632.

[0128] Negation detector 642 may detect negations in answer 640 and / or question 628. For example, if question 628 include a negation, such as “what should I avoid?” or “what is the worst thing on the menu?” Negation detector 642 may determine that a negation is present in question 628. As another example, if answer 640 includes a negation, such as “based on your nut allergy, avoid ...” or “based on your health, do not ...” Negation detector 642 may determine that a negation is present in question 628. Negation detectorPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO31642 may output negation 644 indicative of whether there is a negation present in question 628 and / or answer 640.

[0129] Video data storage 646 data may include videos related to the scene (e.g., scene 506) and / or objects in the scene (e.g., object(s) 508). The videos may be generated by other users, producers, marketers, etc. of objects. The videos may be reviews, instructional videos, demonstrations, etc. The device (e.g.. XR device 504) may remotely access video data storage 646 to obtain video metadata 648. Video metadata 648 may include titles, labels, indices, descriptors of the video data stored by video data storage 646.

[0130] Video RAG 650 may process answer 640 and video metadata 648 to select video information 652. Video RAG 650 may be an example of RAG system 706 of FIG. 7. For example, video RAG 650 may include an encoder and may use the encoder to encode answer 640 to generate features. Similarly, video RAG 650 may use the encoder to encode video metadata 648 to generate various features (e.g., one set of features for each video described by video metadata 648). Video RAG 650 may compare the features based on answer 640 with the features based on video metadata 648 to determine a similarity between answer 640 and various metadata of video metadata 648.

[0131] Video information 652 may be, or may include, indices of various videos selected by video RAG 650. Additionally, video information 652 may include scores associated with the videos selected by video RAG 650. For example, video information 652 may include indications of which videos stored by video data storage 646 are relevant to answer 640.

[0132] Answer selector 654 may select one or more videos stored by video data storage 646 based on video information 652. In some aspects, answer selector 654 may select a highest scoring video (e.g., as indicated by video information 652).

[0133] In cases, in which negation detector 642 determine a negation in question 628 and / or answer 640, answer selector 654 may determine that the answer provided to a user will not include a video. For example, if negation 644 indicates a negation is present in answer 640 and / or question 628, answer selector 654 may determine to provide answer 640 to the user without a video.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO32

[0134] In some case (e.g., cases in which answer selector 654 determines to not provide a video), answer selector 654 may provide answer 640 to the user (e.g., user 502) via speakers 660 of the device (e.g., XR device 504). In such cases, answer selector 654 may provide answer 640 to vocalizer 656 and vocalizer 656 may generate audio answer 658 (e.g., a vocalization of answer 640). Speakers 660 may produce audio answer 658.

[0135] In some cases, (e.g., cases in which answer selector 654 determines to provide video), answer selector 654 may provide visual answer 662 to the user device. Visual answer 662 may include image data, video data, and / or an index of the selected video data. The user device may display the image data and / or the video data (e.g., including playing audio accompanying the video at speakers 660). Additionally or alternatively, the device may access video data 666 via video data storage 646 and play video data 666.

[0136] Image sensor 602. microphone 622, speakers 660, and display 664 may be included the device (e.g., XR device 504). In some cases, object detector 606, ROI determiner 610, scene RAG 618, text generator 626, user RAG 634, LXM 638, negation detector 642, video RAG 650, answer selector 654, and / or vocalizer 656 may be included in the device. In other cases, one or more of object detector 606, ROI determiner 610, scene RAG 618, text generator 626, user RAG 634, LXM 638, negation detector 642, video RAG 650, answer selector 654, and / or vocalizer 656 may be included in a companion device (such as companion device 208 of FIG. 2) or in a remote computing device (such as processing device 206 of FIG. 2).

[0137] As an example of contemplated operations of system 600, a user (e.g., user 502) may ask system 600 for recommendation regarding a class of products. For example, the user may say “can you recommend a coffee that is available in this shop?” System 600 may generate a personalized recommendation that overlaps with the user's personal preferences and constraints and is also available in the shop. Additionally, system 600 may provide a video, if applicable.

[0138] Table 1 includes examples of questions, user data, and user information. For example, Table 2 includes examples of inputs and outputs of user RAG 634. The questions in the first column of table 1 are examples of question 628. The user profile information in the second column are examples of user data 632. The RAG output in the third column are examples of user information 636 (e.g., information that user RAG 634 may determine is relevant to the user query' of the first column of Table 1).Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO33Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO34Table 1

[0139] Table 2 includes examples of detected objects, user data, and scene information. For example, Table 2 includes examples of inputs and outputs of scene RAG 618. The first column of table 2 includes examples of detected objects 608. Additionally, the first column of table 2 also includes examples of user data 632. The second column of table 2 includes examples of scene information 620 (e.g., information that scene RAG 618 may determine is relevant to a question, omitted in table 2, based on detected objects 608 and user data 632).Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO35Table 2

[0140] Table 3 includes examples of answers, video names, and selected video names. For example. Table 3 includes examples of inputs and outputs of video RAG 650. The answers in the first column of Table 3 are examples of answers 640. The video names in the second column of table 3 are examples of names of videos for which video RAG 650 may search (e.g., in an index of videos of video data storage 646). The video names in the third column of Table 3 are examples of names of high-scoring videos selected by video RAG 650.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO36Table 3

[0141] Table 4 includes examples of questions and negation detections. For example, the first column of Table 4 includes examples of question 628. The second column of Table 4 includes examples of negation detections, where “TRUE” indicates a negation is detected in the corresponding question and where “FALSE” indicates that a negation is not detected in the corresponding question.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO37"Table 4

[0142] In some aspects, negation detector 642 may determine negations based on a word list. For example, negation detector 642 may determine negations in questions 628 based on words like: Not , Shouldn't, Cannot, don't, not recommended, Unsafe , risky, avoid, inadvisable, refrain. Additionally, negation detector 642 may determine negations in answers 640 based on words like: Avoid, Avoiding, Don't, Not, Caution, alternative, modify, modifying.

[0143] FIG. 7 is a block diagram of an example system 700 that may generate output 728 based on input 702 and input 704, according to various aspects of the present disclosure. System 700 includes a RAG system 706 that may be used to generate an output 722 based on input 702 and input 704. Additionally, sy stem 700 includes an LXM 726 that may generate output 728 based on input 724 and output 722.

[0144] System 700 may obtain an input 702 and an input 704. Input 702 and input 704 may be relevant to a current context and / or a question to be presented to LXM 726. InputPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO38702 and input 704 are examples of inputs. System 700 may process any number of inputs. Additionally, input 702 and input 704 may be of any format, including, as examples, text, audio data, image data, video data, etc.

[0145] In some aspects, a chunker 708 may divide input 702 into chunks 710 and divide input 704 into chunks 712. Various chunks of chunks 710 and chunks 712 may have similar sizes.

[0146] Encoder 714 may encoder chunks 710 to generate features 716. Additionally, encoder 714 may encode chunks 712 to generate features 718. Features 716 may be a dense feature representation of input 702. Features 718 may be a dense feature representation of input 704. Features 716 and features 718 may ben-dimensional vectors.

[0147] Comparer 720 may compare features 716 to features 718 and determine output 722 based on the comparison. For example, comparer 720 may determine an L2 distance between features 716 and features 718. Comparer 720 may determine portions of input 702 that are most relevant to input 704 based on the similarity between features 716 and features 718. Comparer 720 may output 722. Output 722 may represent portions of input 702 that are most relevant to input 704.

[0148] LXM 726 may process input 724 to generate output 728. Additionally, system 700 may provide output 722 to LXM 726 as an input, such that LXM 726 generates output 728 based on input 724 and comparer 720. In some aspects, system 700 may format output 722 and input 724 to generate a prompt for LXM 726.

[0149] In some aspects, input 704 may be the same as, or may be substantially similar to, input 724. For example, system 700 may generate output 722 such that output 722 represents a portion of input 702 that is relevant to input 724. Further, system 700 may cause LXM 726 to process input 724 and output 722 (e.g., the portion of input 702 relevant to input 724) to generate output 728.

[0150] As mentioned previously, scene RAG 618 may be an example of RAG system 706. For example, detected objects 608, scene data 616, and / or question 628 are examples of inputs (e.g., input 702 and input 704). Further, scene information 620 is an example of output 722.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO39

[0151] Additionally, as mentioned previously, user RAG 634 may be an example of RAG system 706. For example, user data 632 and / or question 628 are examples of inputs (e.g., input 702 and input 704). Further, user information 636 is an example of output 722.

[0152] Additionally, as mentioned previously, video RAG 650 may be an example of RAG system 706. For example, video metadata 648 and / or answer 640 are examples of inputs (e.g., input 702 and input 704). Further, video information 652 is an example of output 722.

[0153] FIG. 8 is a block diagram illustrating an example system 800 for generating an answer 818 to respond to a question 802, according to various aspects of the present disclosure. System 800 represents an example manner in which system 600 may process a question to generate an answer. For example, system 800 represents a timing that system 600 may use in processing data to generate answer 640.

[0154] Question 802 is an example of question 628. User RAG 806 is an example of user RAG 634. User data 804 is an example of user data 632. User information 808 is an example of user information 636. Scene RAG 812 is an example of scene RAG 618. Scene data 810 is an example of scene data 616. Scene information 814 is an example of scene information 620. LXM 816 is an example of LXM 638. Answer 818 is an example of answer 640.

[0155] According to the example of system 800, system 800 may process question 802 and user data 804 using user RAG 806 to generate user information 808. Subsequently, scene RAG 812 may process scene data 810 and, in some cases, user information 808 and / or question 802, to generate scene information 814. Subsequently, LXM 816 may process question 802, user information 808, and scene information 814 to generate answer 818.

[0156] FIG. 9 is a block diagram illustrating an example system 900 for generating an answer 918 to respond to a question 902, according to various aspects of the present disclosure. System 900 represents an example manner in which system 600 may process a question to generate an answer. For example, system 900 represents a timing that system 600 may use in processing data to generate answer 640.

[0157] Question 902 is an example of question 628. User RAG 906 is an example of user RAG 634. User data 904 is an example of user data 632. User information 908 is an Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO40example of user information 636. Scene RAG 912 is an example of scene RAG 618. Scene data 910 is an example of scene data 616. Scene information 914 is an example of scene information 620. LXM 916 is an example of LXM 638. Answer 918 is an example of answer 640.

[0158] According to the example of system 900, system 900 may process question 902 and user data 904 using user RAG 906 to generate user information 908. At substantially the same time, scene RAG 912 may process scene data 910 and / or question 902, to generate scene information 914. Subsequently, LXM 916 may process question 902, user information 908 and scene information 914 to generate answer 918.

[0159] One advantage of the example timing illustrated by system 900 over the example timing illustrated by system 800 is that user RAG 906 and scene RAG 912 may operate in parallel, for example, at substantially the same time. Operating user RAG 906 and scene RAG 912 at substantially the same time may reduce a run time of system 900 as compared with system 800. Thus, if system 600 operates according to the timing illustrated by system 900, system 600 may operate more quickly than system 600 operates according to the timing illustrated by system 800.

[0160] FIG. 10 is a flow diagram illustrating an example process 1000 that an answer selector (e.g., answer selector 654) may use to select an answer from among a plurality of answers, according to various aspects of the present disclosure. For example, a RAG (e.g., video RAG 650) may provide an indication of two or more videos that may be relevant to a question (e.g., answer 640) to an answer selector (e.g., answer selector 654). The indications of the two or more videos may include relevancy scores. As an example, the highest relevancy score may be referred to as “topi.” The second highest relevancy score may be referred to as “top2.”

[0161] At decision block 1002, if the highest relevancy score “topi” exceeds a threshold (e.g., 0.8 on a scale betw een 0 and 1), answer selector 654 may select the video corresponding to the highest relevancy score as the answ er to the question. If the highest relevancy score does not exceed the threshold, process 1000 may proceed to decision block 1004.

[0162] At decision block 1004 a ratio betw een the relevancy score of the video with the highest relevancy score may be compared to the relevancy score to the video with thePolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO41second-highest relevancy score. If the ratio exceeds a threshold (e.g., 1.1), the video corresponding to the highest relevancy score is selected as the answer to the question (e.g., at block 1006). If the ratio does not exceed the threshold, process 1000 may proceed to block 1008 and neither of the first or the second video may be selected as the answer.

[0163] FIG. 11 is a flow diagram illustrating an example process 1100 for generating answers to questions, in accordance with aspects of the present disclosure. One or more operations of process 1100 may be performed by a computing device (or apparatus) or a component (e.g., a chipset, codec, etc.) of the computing device. The computing device may be a mobile device (e.g., a mobile phone), a network-connected wearable such as a watch, an extended reality (XR) device such as a virtual reality (VR) device or augmented reality (AR) device, a vehicle or component or system of a vehicle, a desktop computing device, a tablet computing device, a server computer, a robotic device, and / or any other computing device with the resource capabilities to perform the one or more operations of process 1100. The one or more operations of process 1100 may be implemented as software components that are executed and run on one or more processors.

[0164] At block 1102, a computing device (or one or more components thereof) may detect an object in a scene of a user. For example, object detector 606 may detect an object in a scene based on image data 604 (e.g., images of the scene).

[0165] At block 1104, the computing device (or one or more components thereof) may obtain scene data associated with the scene. For example, system 600 may obtain scene data 616. Scene data 616 may be associated with the scene represented by image data 604.

[0166] In some aspects, the computing device (or one or more components thereof) may detect at least one of a quick-response (QR) code, a name associated with the scene, or a recognizable landmark in the image of the scene; identify the scene based on at least one of the quick-response (QR) code, the name associated with the scene, or the recognizable landmark; and identify a data storage associated with the scene based on identification of the scene. The scene data may be obtained from the data storage associated with the scene. For example, object detector 606 may detect a QR code in image data 604, detect a name in image data 604, or recognize a recognizable landmark in image data 604. System 600 may identify the scene based on the QR code, the name, and / or the recognizable landmark. System 600 may identify scene data storage 614 based Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO42on the identified scene. System 600 may obtain scene data 616 from scene data storage 614.

[0167] At block 1106, the computing device (or one or more components thereof) may generate scene information based on the detected object and the scene data. For example, scene RAG 618 may generate scene information 620 based on detected object 608 and scene data 616.

[0168] In some aspects, the computing device (or one or more components thereof) may generate the scene information by processing indications of the detected object and the scene data using a retrieval-augmented-generation (RAG) model. For example, scene RAG 618 may generate scene information 620 by processing detected object 608 and scene data 616.

[0169] At block 1108, the computing device (or one or more components thereof) may identify a region of interest (ROI) of an image of the scene. For example, ROI determiner 610 may determine ROI 612 based on image data 604.

[0170] In some aspects, the ROI may be determined based on a gaze of the user; a center of the image; text detected in the image; or objects detected in the image. For example, ROI determiner 610 may determine ROI 612 based on a gaze of the user, a center of the image, text detected in the image, and / or obj ects detected in the image.

[0171] In some aspects, the image may be obtained by a scene-facing camera of a headmounted device. For example, image sensor 602 may capture image data 604. Image sensor 602 may be. or may include, a scene-facing camera of a HMD.

[0172] At block 1110, the computing device (or one or more components thereof) may generate user information based on a question and user data associated with the user. For example, user RAG 634 may generate user information 636 based on user data 632 and question 628.

[0173] In some aspects, the user information is generated by processing the question and the user data using a retrieval-augmented-generation (RAG) model. For example, user RAG 634 may generate user information 636 by processing question 628 and user data 632.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO43

[0174] In some aspects, the user data is obtained from a data storage associated with the user. For example, system 600 may obtain user data 632 from user data storage 630. User data storage 630 may be associated wi th the user.

[0175] In some aspects, to obtain the question, the computing device (or one or more components thereof) may record the question as spoken by the user; and convert the spoken question into text. For example, microphone 622 may capture the question as spoken by a user and text generator 626 may convert audio data 624 to question 628.

[0176] At block 1112, the computing device (or one or more components thereof) may process the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer. For example, LXM 638 may process question 628, image data 604, ROI 612, scene information 620, and user information 636 to generate answer 640.

[0177] In some aspects, to process the question, the image of the scene, the ROI, the scene information, and the user information using the machine-learning model, the computing device (or one or more components thereof) may format the question, the image of the scene, the ROI, the scene information, and the user information as a prompt for the machine-learning model. For example, system 600 may format question 628, image data 604, ROI 612, scene information 620 and user information 636 as an input to LXM 638.

[0178] In some aspects, the machine-learning model may be, or may include, a large multimodal machine-learning model.

[0179] At block 1114, the computing device (or one or more components thereof) may identify a video from among a plurality of videos that is relevant to the answer. For example, video RAG 650 may identify a video from among a plurality of videos stored by video data storage 646. Video RAG 650 may select the video based on determining that the selected video is relevant to answer 640.

[0180] In some aspects, to identify the video from among the plurality of videos, the computing device (or one or more components thereof) may process the answer and at least one of titles or labels of the plurality of videos using a retrieval-augmented-generation (RAG) model. For example, video RAG 650 may process answer 640 and video metadata 648.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO44

[0181] At block 1116, the computing device (or one or more components thereof) may provide at least one of the answer or the video to the user. For example, system 600 may provide answer 640 to a user. For instance, system 600 may provide audio answer 658 to the user or visual answer 662 to the user.

[0182] In some aspects, the computing device (or one or more components thereof) may detect negations in at least one of the question or the answer; and in response to detecting a negation in at least one of the question or the answer, determine to provide the answer to the user. For example, negation detector 642 may determine negation 644 based on answer 640 and / or question 628. In some cases, answer selector 654 may determine to provide answer 640 to the user rather than video information 652 based on negation 644.

[0183] In some aspects, the computing device (or one or more components thereof) may detect negations in at least one of the question or the answer; and in response to not detecting a negation in at least one of the question or the answer, determine to provide the video to the user. For example, negation detector 642 may determine negation 644 based on answer 640 and / or question 628. In some cases, answer selector 654 may determine to provide visual answer 662 to the user.

[0184] In some aspects, to provide the answer to the user, the computing device (or one or more components thereof) may vocalize the answer using a speaker of a head-mounted device. For example, vocalizer 656 may vocalize answer 640 to generate audio answer 658. Further speakers 660 may output audio answer 658.

[0185] In some aspects, to provide the video to the user, the computing device (or one or more components thereof) may play the video using a display of a head-mounted device. For example, display 664 may display visual answer 662.

[0186] In some examples, as noted previously, the methods described herein (e.g., process 1100 of FIG. 11, and / or other methods described herein) can be performed, in whole or in part, by a computing device or apparatus. In one example, one or more of the methods can be performed by XR system 100 of FIG. 1, XR device 102 of FIG. 1, XR system 200 of FIG. 2, display device 204 of FIG. 2, processing device 206 of FIG. 2, companion device 208 of FIG. 2, XR system 300 of FIG. 3, display 304 of FIG. 3, XR system 400 of FIG. 4, XR device 504 of FIG. 5, system 600 of FIG. 6, system 700 of FIG.7, system 800 of FIG. 8, system 900 of FIG. 9, or by another system or device. In anotherPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO45example, one or more of the methods (e.g., process 1100, and / or other methods described herein) can be performed, in whole or in part, by the computing-device architecture 1700 shown in FIG. 17. For instance, a computing device with the computing-device architecture 1700 shown in FIG. 17 can include, or be included in, the components of the XR system 100 of FIG. 1, XR device 102 of FIG. 1, XR system 200 of FIG. 2, display device 204 of FIG. 2, processing device 206 of FIG. 2, companion device 208 of FIG. 2, XR system 300 of FIG. 3, display 304 of FIG. 3, XR system 400 of FIG. 4, XR device 504 of FIG. 5, system 600 of FIG. 6, system 700 of FIG. 7, system 800 of FIG. 8, system 900 of FIG. 9 and can implement the operations of process 1100, and / or other process described herein. In some cases, the computing device or apparatus can include various components, such as one or more input devices, one or more output devices, one or more processors, one or more microprocessors, one or more microcomputers, one or more cameras, one or more sensors, and / or other component(s) that are configured to carry out the steps of processes described herein. In some examples, the computing device can include a display, a network interface configured to communicate and / or receive the data, any combination thereof, and / or other component(s). The network interface can be configured to communicate and / or receive Internet Protocol (IP) based data or other type of data.

[0187] The components of the computing device can be implemented in circuitry. For example, the components can include and / or can be implemented using electronic circuits or other electronic hardware, which can include one or more programmable electronic circuits (e.g., microprocessors, graphics processing units (GPUs), digital signal processors (DSPs), central processing units (CPUs), and / or other suitable electronic circuits), and / or can include and / or be implemented using computer software, firmware, or any combination thereof, to perform the various operations described herein.

[0188] Process 1100, and / or other process described herein are illustrated as logical flow diagrams, the operation of which represents a sequence of operations that can be implemented in hardware, computer instructions, or a combination thereof. In the context of computer instructions, the operations represent computer-executable instructions stored on one or more computer-readable storage media that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions include routines, programs, objects, components, data structures, and the likePolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO46that perform particular functions or implement particular data types. The order in which the operations are described is not intended to be construed as a limitation, and any number of the described operations can be combined in any order and / or in parallel to implement the processes.

[0189] Additionally, process 1100, and / or other process described herein can be performed under the control of one or more computer systems configured with executable instructions and can be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) executing collectively on one or more processors, by hardware, or combinations thereof. As noted above, the code can be stored on a computer-readable or machine-readable storage medium, for example, in the form of a computer program comprising a plurality of instructions executable by one or more processors. The computer-readable or machine-readable storage medium can be non-transitory.

[0190] As noted above, various aspects of the present disclosure can use machinelearning models or systems.

[0191] FIG. 12 is an illustrative example of aneural network 1200 (e.g., a deep-learning neural network) that can be used to implement machine-learning based feature segmentation, implicit-neural -representation generation, rendering, classification, object detection, image recognition (e.g., face recognition, object recognition, scene recognition, etc.), feature extraction, authentication, gaze detection, gaze prediction, and / or automation. For example, neural network 1200 may be an example of, or can implement, object detector 606, ROI determiner 610, scene RAG 618, text generator 626, user RAG 634, LXM 638, negation detector 642, video RAG 650, and / or vocalizer 656, of FIG. 6, chunker 708, encoder 714, comparer 720, RAG system 706, and / or LXM 726 of FIG. 7, user RAG 806, scene RAG 812, and / or LXM 816 of FIG. 8, user RAG 906, scene RAG 912, and / or LXM 816 of FIG. 8.

[0192] An input layer 1202 includes input data. In one illustrative example, input layer 1202 can include data representing image data, text data, video data, audio data, etc. Neural network 1200 includes multiple hidden layers, for example, hidden layers 1206a, 1206b, through 1206n. The hidden layers 1206a, 1206b, through hidden layer 1206n include “n’’ number of hidden layers, where “n” is an integer greater than or equal to one. The number of hidden layers can be made to include as many layers as needed for the Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO47given application. Neural network 1200 further includes an output layer 1204 that provides an output resulting from the processing performed by the hidden layers 1206a, 1206b, through 1206n. In one illustrative example, output layer 1204 can provide feature vectors, numerical data, image data, text data, video data, audio data, etc.

[0193] Neural network 1200 may be, or may include, a multi-layer neural network of interconnected nodes. Each node can represent a piece of information. Information associated with the nodes is shared among the different layers and each layer retains information as information is processed. In some cases, neural network 1200 can include a feed-forward network, in which case there are no feedback connections where outputs of the network are fed back into itself. In some cases, neural network 1200 can include a recunent neural network, which can have loops that allow information to be carried across nodes while reading in input.

[0194] Information can be exchanged between nodes through node-to-node interconnections between the various layers. Nodes of input layer 1202 can activate a set of nodes in the first hidden layer 1206a. For example, as show n, each of the input nodes of input layer 1202 is connected to each of the nodes of the first hidden layer 1206a. The nodes of first hidden layer 1206a can transform the information of each input node by applying activation functions to the input node information. The information derived from the transformation can then be passed to and can activate the nodes of the next hidden layer 1206b, which can perform their own designated functions. Example functions include convolutional, up-sampling, data transformation, and / or any other suitable functions. The output of the hidden layer 1206b can then activate nodes of the next hidden layer, and so on. The output of the last hidden layer 1206n can activate one or more nodes of the output layer 1204, at which an output is provided. In some cases, while nodes (e.g., node 1208) in neural network 1200 are shown as having multiple output lines, a node has a single output and all lines show n as being output from a node represent the same output value.

[0195] In some cases, each node or interconnection between nodes can have a weight that is a set of parameters derived from the training of neural network 1200. Once neural network 1200 is trained, it can be referred to as a trained neural network, which can be used to perform one or more operations. For example, an interconnection between nodes can represent a piece of information learned about the interconnected nodes. The Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO48interconnection can have a tunable numeric weight that can be tuned (e.g., based on a training dataset), allowing neural network 1200 to be adaptive to inputs and able to leam as more and more data is processed.

[0196] Neural netw ork 1200 may be pre-trained to process the features from the data in the input layer 1202 using the different hidden layers 1206a, 1206b, through 1206n in order to provide the output through the output layer 1204. In an example in which neural network 1200 is used to identify features in images, neural network 1200 can be trained using training data that includes both images and labels, as described above. For instance, training images can be input into the network, with each training image having a label indicating the features in the images (for the feature-segmentation machine-learning system) or a label indicating classes of an activity in each image. In one example using object classification for illustrative purposes, a training image can include an image of a number 2, in which case the label for the image can be [00 1 0 0000 00],

[0197] In some cases, neural netw ork 1200 can adjust the w eights of the nodes using a training process called backpropagation. As noted above, a backpropagation process can include a forward pass, a loss function, a backward pass, and a weight update. The forward pass, loss function, backward pass, and parameter update are performed for one training iteration. The process can be repeated for a certain number of iterations for each set of training images until neural network 1200 is trained w ell enough so that the w eights of the layers are accurately tuned.

[0198] For the example of identifying objects in images, the forw ard pass can include passing a training image through neural network 1200. The weights are initially randomized before neural network 1200 is trained. As an illustrative example, an image can include an array of numbers representing the pixels of the image. Each number in the array can include a value from 0 to 255 describing the pixel intensity7at that position in the array. In one example, the array can include a 28 x 28 x 3 array of numbers with 28 rows and 28 columns of pixels and 3 color components (such as red, green, and blue, or luma and two chroma components, or the like).

[0199] As noted above, for a first training iteration for neural network 1200, the output will likely include values that do not give preference to any particular class due to the weights being randomly selected at initialization. For example, if the output is a vector with probabilities that the object includes different classes, the probability value for each Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO49of the different classes can be equal or at least very similar (e.g., for ten possible classes, each class can have a probability value of 0.1). With the initial weights, neural network 1200 is unable to determine low-level features and thus cannot make an accurate determination of what the classification of the object might be. A loss function can be used to analyze error in the output. Any suitable loss function definition can be used, such as a cross-entropy loss. Another example of a loss function includes the mean squared error (MSE), defined as Etotai= 2“ target — output2. The loss can be set to be equal to the value of Etotal.

[0200] The loss (or error) will be high for the first training images since the actual values will be much different than the predicted output. The goal of training is to minimize the amount of loss so that the predicted output is the same as the training label. Neural network 1200 can perform a backward pass by determining which inputs (weights) most contributed to the loss of the network and can adjust the weights so that the loss decreases and is eventually minimized. A derivative of the loss with respect to the weights (denoted where W are the weights at a particular layer) can be computed to determine the weights that contributed most to the loss of the network. After the derivative is computed, a weight update can be performed by updating all the weights of the filters. For example, the weights can be updated so that they change in the opposite direction of the gradient. The weight update can be denoted as w — w±— pwhere w denotes a weight, w, denotes the initial weight, and q denotes a learning rate. The learning rate can be set to any suitable value, with a high learning rate including larger weight updates and a lower value indicating smaller weight updates.

[0201] Neural network 1200 can include any suitable deep network. One example includes a convolutional neural network (CNN), which includes an input layer and an output layer, with multiple hidden layers between the input and out layers. The hidden layers of a CNN include a series of convolutional, nonlinear, pooling (for downsampling), and fully connected layers. Neural network 1200 can include any other deep network other than a CNN, such as an autoencoder, a deep belief nets (DBNs), a Recurrent Neural Networks (RNNs). among others.

[0202] FIG. 13 is an illustrative example of a convolutional neural network (CNN) 1300. The input layer 1302 of the CNN 1300 includes data representing an image or Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO50frame. For example, the data can include an array of numbers representing the pixels of the image, with each number in the array including a value from 0 to 255 describing the pixel intensity7at that position in the array. Using the previous example from above, the array can include a 28 x 28 x 3 array of numbers with 28 rows and 28 columns of pixels and 3 color components (e.g., red, green, and blue, or luma and two chroma components, or the like). The image can be passed through a convolutional hidden layer 1304, an optional non-linear activation layer, a pooling hidden layer 1306, and fully connected layer 1308 (which fully connected layer 1308 can be hidden) to get an output at the output layer 1310. While only one of each hidden layer is shown in FIG. 13, one of ordinary skill will appreciate that multiple convolutional hidden layers, non-linear layers, pooling hidden layers, and / or fully connected layers can be included in the CNN 1300. As previously described, the output can indicate a single class of an object or can include a probability of classes that best describe the object in the image.

[0203] The first layer of the CNN 1300 can be the convolutional hidden layer 1304. The convolutional hidden layer 1304 can analyze image data of the input layer 1302. Each node of the convolutional hidden layer 1304 is connected to a region of nodes (pixels) of the input image called a receptive field. The convolutional hidden layer 1304 can be considered as one or more filters (each filter corresponding to a different activation or feature map), with each convolutional iteration of a filter being a node or neuron of the convolutional hidden layer 1304. For example, the region of the input image that a filter covers at each convolutional iteration would be the receptive field for the filter. In one illustrative example, if the input image includes a 28x28 array, and each filter (and corresponding receptive field) is a 5x5 array, then there will be 24x24 nodes in the convolutional hidden layer 1304. Each connection between anode and a receptive field for that node learns a weight and, in some cases, an overall bias such that each node learns to analyze its particular local receptive field in the input image. Each node of the convolutional hidden layer 1304 will have the same weights and bias (called a shared weight and a shared bias). For example, the filter has an array of weights (numbers) and the same depth as the input. A filter will have a depth of 3 for an image frame example (according to three color components of the input image). An illustrative example size of the filter array is 5 x 5 x 3, corresponding to a size of the receptive field of a node.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO51

[0204] The convolutional nature of the convolutional hidden layer 1304 is due to each node of the convolutional layer being applied to its corresponding receptive field. For example, a filter of the convolutional hidden layer 1304 can begin in the top-left comer of the input image array and can convolve around the input image. As noted above, each convolutional iteration of the filter can be considered a node or neuron of the convolutional hidden layer 1304. At each convolutional iteration, the values of the filter are multiplied with a corresponding number of the original pixel values of the image (e.g., the 5x5 filter array is multiplied by a 5x5 array of input pixel values at the top-left comer of the input image array). The multiplications from each convolutional iteration can be summed together to obtain a total sum for that iteration or node. The process is next continued at a next location in the input image according to the receptive field of a next node in the convolutional hidden layer 1304. For example, a filter can be moved by a step amount (referred to as a stride) to the next receptive field. The stride can be set to 1 or any other suitable amount. For example, if the stride is set to 1, the filter will be moved to the right by 1 pixel at each convolutional iteration. Processing the filter at each unique location of the input volume produces a number representing the filter results for that location, resulting in a total sum value being determined for each node of the convolutional hidden layer 1304.

[0205] The mapping from the input layer to the convolutional hidden layer 1304 is referred to as an activation map (or feature map). The activation map includes a value for each node representing the filter results at each location of the input volume. The activation map can include an array that includes the various total sum values resulting from each iteration of the filter on the input volume. For example, the activation map will include a 24 x 24 array if a 5 x 5 filter is applied to each pixel (a stride of 1) of a 28 x 28 input image. The convolutional hidden layer 1304 can include several activation maps in order to identify multiple features in an image. The example shown in FIG. 13 includes three activation maps. Using three activation maps, the convolutional hidden layer 1304 can detect three different kinds of features, with each feature being detectable across the entire image.

[0206] In some examples, a non-linear hidden layer can be applied after the convolutional hidden layer 1304. The non-linear layer can be used to introduce nonlinearity to a system that has been computing linear operations. One illustrative examplePolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO52of a non-linear layer is a rectified linear unit (ReLU) layer. A ReLU layer can apply the function f(x) = max(0, x) to all of the values in the input volume, which changes all the negative activations to 0. The ReLU can thus increase the non-linear properties of the CNN 1300 without affecting the receptive fields of the convolutional hidden layer 1304.

[0207] The pooling hidden layer 1306 can be applied after the convolutional hidden layer 1304 (and after the non-linear hidden layer when used). The pooling hidden layer 1306 is used to simplify the information in the output from the convolutional hidden layer 1304. For example, the pooling hidden layer 1306 can take each activation map output from the convolutional hidden layer 1304 and generates a condensed activation map (or feature map) using a pooling function. Max-pooling is one example of a function performed by a pooling hidden layer. Other forms of pooling functions be used by the pooling hidden layer 1306, such as average pooling, L2-norm pooling, or other suitable pooling functions. A pooling function (e.g., a max-pooling filter, an L2-norm filter, or other suitable pooling filter) is applied to each activation map included in the convolutional hidden layer 1304. In the example shown in FIG. 13, three pooling filters are used for the three activation maps in the convolutional hidden layer 1304.

[0208] In some examples, max-pooling can be used by applying a max-pooling filter (e.g., having a size of 2x2) with a stride (e g., equal to a dimension of the filter, such as a stride of 2) to an activation map output from the convolutional hidden layer 1304. The output from a max-pooling filter includes the maximum number in every sub-region that the filter convolves around. Using a 2x2 filter as an example, each unit in the pooling layer can summarize a region of 2x2 nodes in the previous layer (with each node being a value in the activation map). For example, four values (nodes) in an activation map will be analyzed by a 2x2 max-pooling filter at each iteration of the filter, with the maximum value from the four values being output as the "max" value. If such a max-pooling filter is applied to an activation filter from the convolutional hidden layer 1304 having a dimension of 24x24 nodes, the output from the pooling hidden layer 1306 will be an array of 12x12 nodes.

[0209] In some examples, an L2-norm pooling filter could also be used. The L2-norm pooling filter includes computing the square root of the sum of the squares of the values in the 2x2 region (or other suitable region) of an activation map (instead of computingPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO53the maximum values as is done in max-pooling) and using the computed values as an output.

[0210] The pooling function (e.g., max-pooling, L2-norm pooling, or other pooling function) determines whether a given feature is found anywhere in a region of the image and discards the exact positional information. This can be done without affecting results of the feature detection because, once a feature has been found, the exact location of the feature is not as important as its approximate location relative to other features. Maxpooling (as well as other pooling methods) offer the benefit that there are many fewer pooled features, thus reducing the number of parameters needed in later layers of the CNN 1300.

[0211] The final layer of connections in the network is a fully -connected layer that connects every node from the pooling hidden layer 1306 to every one of the output nodes in the outputlayer 1310. Using the example above, the input layer includes 28 x 28 nodes encoding the pixel intensities of the input image, the convolutional hidden layer 1304 includes 3x24x24 hidden feature nodes based on application of a 5x5 local receptive field (for the filters) to three activation maps, and the pooling hidden layer 1306 includes a layer of 3x 12x 12 hidden feature nodes based on application of max-pooling filter to 2x2 regions across each of the three feature maps. Extending this example, the output layer 1310 can include ten output nodes. In such an example, every' node of the 3x12x12 pooling hidden layer 1306 is connected to every node of the output layer 1310.

[0212] The fully connected layer 1308 can obtain the output of the previous pooling hidden layer 1306 (which should represent the activation maps of high-level features) and determines the features that most correlate to a particular class. For example, the fully connected layer 1308 can determine the high-level features that most strongly correlate to a particular class and can include weights (nodes) for the high-level features. A product can be computed between the weights of the fully connected layer 1308 and the pooling hidden layer 1306 to obtain probabilities for the different classes. For example, if the CNN 1300 is being used to predict that an obj ect in an image is a person, high values will be present in the activation maps that represent high-level features of people (e.g., two legs are present, a face is present at the top of the object, two eyes are present at the top left and top right of the face, a nose is present in the middle of the face, a mouth is present at the bottom of the face, and / or other features common for a person).Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO54

[0213] In some examples, the output from the output layer 1310 can include an M-dimensional vector (in the prior example, M=10). M indicates the number of classes that the CNN 1300 has to choose from when classifying the object in the image. Other example outputs can also be provided. Each number in the M-dimensional vector can represent the probability the object is of a certain class. In one illustrative example, if a 10-dimensional output vector represents ten different classes of objects is [00 0.05 0.80 0.15 0000], the vector indicates that there is a 5% probability that the image is the third class of object (e.g., a dog), an 80% probability that the image is the fourth class of object (e.g., a human), and a 15% probability that the image is the sixth class of object (e.g.. a kangaroo). The probability for a class can be considered a confidence level that the object is part of that class.

[0214] FIG. 14 includes an example machine-learning model 1400 that may be used in various aspects of the present disclosure. For example, any or all of LXM 638 of FIG. 6, LXM 726 of FIG. 7, LXM 816 of FIG. 8, and / or LXM 916 of FIG. 9 may be examples of machine-learning model 1400.

[0215] Machine-learning model 1400 is an example of a generative response engine. Generative response engines are commonly referred to as Generative Al. Generative response engines can receive an input prompt (e.g., input 1406) and generate content (e g., output 1408) based on the prompt. Generative Pre-trained Transformers (GPTs), diffusion models, and diffusion-transformer models are some non-limiting examples of generative response engines.

[0216] Machine-learning model 1400 includes a predictive output-generation engine 1402 and Output validation engine 1404. Predictive output-generation engine 1402 may analyze input 1406 and identify relevant patterns and associations based on data on which predictive output-generation engine 1402 was trained. Further, predictive outputgeneration engine 1402 may predict a sequence of words that are the most likely continuation of input 1406. By iteratively predicting next words, predictive outputgeneration engine 1402 may aim to provide a coherent and contextually relevant answer to input 1406. Predictive output-generation engine 1402 may generate responses by sampling from the probability distribution of possible words and sequences, guided by the patterns observed during the training of predictive output-generation engine 1402. In some aspects, predictive output-generation engine 1402 may generate multiple possible Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO55responses before outputting a final one. The multiple responses may be variations that predictive output-generation engine 1402 considers potentially relevant and coherent. Output validation engine 1404 may evaluate the multiple generated responses based on certain criteria. These criteria can include relevance to the prompt, coherence, fluency, and sometimes adherence to specific guidelines or rules, depending on the application. Based on this evaluation, output validation engine 1404 may select a most appropriate response. This selection is typically the one that scores highest on the set criteria, balancing factors like relevance, informativeness, and coherence.

[0217] Input 1406 and / or output 1408 may be, or may include, text, image data, video data, numerical data, etc. For example, machine-learning model 1400 may perform tasks such as, text summarization, text translation, text generation, responding to queries, image description, video description, image generation (e.g., based on text and / or image data), video generation (e.g., based on text and / or image data), image rendering (e.g., based on a 3D model and / or image data), object detection (e.g.. based on image data and / or video data) etc. As such, machine-learning model 1400 may be referred to as a large language model (LLM), a vision-language model (VLM), a multilingual language model (MLLM) a large vision model (LVM), a large multimodal model (LMM), etc.

[0218] FIG. 15 is a block diagram illustrating a multimodal generative ML system 1500 for generating natural language responses based on natural language input from a prompt 1502 andany additional information. A multimodal machine learning system is amachine learning model that receives, processes, and outputs data in multiple forms. For example, the input prompt may include text, images, and audio. Any or all of LXM 638 of FIG. 6, LXM 726 of FIG. 7, LXM 816 of FIG. 8, and / or LXM 916 of FIG. 9 may be examples of multimodal generative ML system 1500.

[0219] For example, the multimodal generative ML system 1500 includes a plurality7of encoders 1504 that are each configured to encode different modes of content (e.g., text, images, audio, etc.) into different tokens within a common embedding space. For example, a text input may be segmented based on different techniques (e g., paragraph, sentence, etc.) and encoded by a text encoder (from the encoders 1504) into tokens. In another example, one or more images can be provided to an image encoder (from the encoders 1504) that extracts features associated with the image and generates tokens representing the visual features. In another example, audio can be provided to an audio Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO56encoder (from the encoders 1504) that extracts features associated with the image and generates tokens representing the audio features. In the case of audio, the audio encoder can identify features that can include formants that characterize resonant frequencies in speech, rhythmic features related to timing and tempo, and harmonic features that describe the relationship between fundamental frequencies and their harmonics.

[0220] The different tokens from the plurality of encoders are provided to combiner 1506. The combiner 1506 can combine the tokens based on the order in which they are presented. For example, the input into the encoder may be an array of primitive values. A primitive value is an immutable data type provided by a programming language and includes values that represent a single piece of data (e.g.. number, string, Boolean, etc.) rather than a complex object or reference. A non-limiting example prompt may include a byte array (e.g., an unsigned 8-byte integer array or uint8array), and another string. The byte array may be audio, images, or other content that can be processed by the encoders 1504. In some aspects, the combiner 1506 is configured to concatenate the different tokens in order based on the array to preserve the semantic order of features and provide the tokens to the generative machine learning model 1508.

[0221] The generative machine learning model 1508 is configured to receive the tokens and generate a natural language response 1512 based on the tokens and the prompt 1502. Generative machine learning model 1508 may include one or more models 1510 (e.g., transformer neural network(s), diffusion model(s), fully connected layer(s). multilayer perceptrons (MLPs), any combination thereof, and / or other models). The one or more models 1510 of the generative machine learning model 1508 are configured to process the tokens and extract different types of features that are relevant to the prompt 1502. For example, the prompt 1502 can be a query' for a particular type of information. The one or more models of the generative machine learning model 1508 can perform different tasks related to the query; such as writing code to perform a particular function, generating an image based on an input image yvith expressed modifications, generate an image without any input image, and so forth.

[0222] The generative machine learning model 1508 may include different components, such as a featurization engine to identify different types of features, an inference engine to identify inferences yvithin the text (e.g., pronoun usage and corresponding disambiguation functions), data retrieval engines (e.g., to identify features Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO57related to a particular concept observed by the generative machine learning model 1508), and so forth. The generative machine learning model 1508 may also include different types of models and engines to synthesize a coherent contextual output, such as to synthesize the input content and information that is responsive to tasks embedded within the text. For example, the generative machine learning model 1508 may include a predictive output engine (not shown) that is configured to generate a sequence of words that is most likely contextually correct and to provide a coherent and contextually relevant answer. For instance, the predictive output generation engine can generate responses by sampling from the probability distribution of possible words and sequences based on patterns observed during training. The generative machine learning model 1508 may also include a predictive output generation engine to generate multiple responses that are potentially relevant and coherent with respect to the prompt 1502. The generative machine learning model 1508 may also include an output validation engine configured to evaluate the generated responses based on certain criteria. Non-limiting examples of criteria to evaluate generated responses include relevance to the prompt, coherence, fluency, and adherence to specific guidelines or rules. Based on the evaluation, the output validation engine may select and output the most appropriate response.

[0223] As noted above, the generative machine learning model 1508 may include various types of models (e.g., machine learning models), such as a transformer. A transformer is a neural network architecture that can be trained to perform one or more natural language processing (NLP) tasks, such as language translation, sentiment analysis, and text summarization. Conventional traditional recurrent neural networks (RNNs) process data in sequence. A transformer or transformer network can process input in parallel and can thus be faster and more efficient than sequential training and processing. In some aspects, a transformer can use a self-attention mechanism (e.g., one or more selfattention layers), which allows the transformer to identify the most relevant parts of the input text or content (e.g., audio or video). In some cases, a transformer can also use a cross-attention mechanism (e.g., one or more cross-attention layers) which uses other content or data to determine the most relevant parts of the input. For example, crossattention mechanisms are useful in sequential content such as a stream of data, such as optical flow, and other computer vision techniques.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO58

[0224] A transformer neural network can include a multi-layer encoder-decoder architecture. For instance, an encoder of the encoder-decoder architecture can receive text as input, convert the input text into a sequence of hidden representations, and capture the meaning of the text at different levels of abstraction. A decoder of the encoder-decoder architecture can then process the representations output from the decoder to generate an output sequence, such as a text translation or a summary. The encoder and decoder can be trained together using supervised learning, unsupervised learning, or a combination of supervised and unsupervised learning techniques, such as maximum likelihood estimation and self-supervised pretraining. Illustrative examples of transformer engines include a BERT model, a Text-to-Text Transfer Transformer (T5), biomedical BERT (BioBERT), scientific BERT (SciBERT), and the SPECTER model for document-level representation learning. In some aspects, multiple transformer engines may be used to generate different tokens.

[0225] In some aspects, the generative machine learning model 1508 may be executed using a neural engine (or multiple neural engines) for on-device execution, such as a neural processing unit (NPU), a neural signal processor (NSP), a digital signal processor (DSP), any combination thereof, and / or other neural engine. The neural engine can include a plurality' of neural processing cores that are configured to parallelize operations associated with neural networks. A neural processing core can include arrays of multiply -accumulate (MAC) units and specialized instructions that are optimized for matrix operations, such as convolution and matrix multiplication. The neural processing core can receive input data and perform matrix transformations and nonlinear activation functions to break down and parallelize matrix operations. The neural processing core can perform tasks such as inference (e.g.. runtime operation of a machine learning model) or training of deep learning models. The neural processing core can accelerate tasks by parallelization of larger computations that can be performed in parallel (e.g., matrix operations associated with neural networks). For instance, the neural engine may perform computer vision tasks such as object recognition. In some cases, the neural engine can be implemented based on various ML libraries such as PyTorch, which interfaces with the compute unified device architecture (CUD A) to parallelize operations.

[0226] In some aspects, the generative machine learning model 1508 may be a small generative model that has fewer parameters, fewer layers, fewer neurons, or a simplerPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO59architecture compared to larger models. A small generative model may not capture the full complexity of the underlying data distribution as effectively as larger models but can still be useful in scenarios where computational resources are limited or where a simpler model is sufficient for the task. Small generative models can also be easier to train and interpret, making them suitable for certain applications. For example, ChatGPT-3.5 has 175 billion parameters that results in a size of 1.4 Terabytes (TB) for a model implemented with double-precision floating point numbers. A smaller model may have a simpler architecture, use fewer parameters (e.g., 10 million), and use less precise numbers (e.g., single-precision floating point numbers) resulting in a size of 38 Megabytes (MB).

[0227] In addition, small models benefit from increased training based on local execution and data specific to a local device and a user of that local device. An additional benefit to small models is increased privacy because the information is not transmitted over the network and only relies on information requested by the user or usage at the local device.

[0228] FIG. 16 is a block diagram of an example transformer in accordance with some aspects of the disclosure. For example, transformer 1600 may be included in any or all of LXM 638 of FIG. 6, LXM 726 of FIG. 7. LXM 816 of FIG. 8, LXM 916 of FIG. 9. machine-learning model 1400 of FIG. 14, and / or multimodal generative ML system 1500 of FIG. 15.

[0229] In a convolutional neural network (CNN) model, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, which makes learning dependencies at different distant positions challenging for a CNN model. A transformer 1600 reduces the operations of learning dependencies by using an encoder 1610 and a decoder 1630 that implement an attention mechanism at different positions of a single sequence to compute a representation of that sequence. An attention function can be described as mapping a query and a set of keyvalue pairs to an output, where the query, keys, values, and output are all vectors. The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility' function of the query' with the corresponding key.

[0230] In one example of a transformer, the encoder 1610 is composed of a stack of six identical layers and each layer has two sub-layers. The first sub-layer is a multi-head selfattention engine 1612, and the second sub-layer is a fully-connected feed-forward Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO60network 1614. A residual connection (not shown) connects around each of the sub-layers followed by normalization.

[0231] In this example transformer 1600, the decoder 1630 is also composed of a stack of six 6 identical layers. The decoder also includes a masked multi-head self-attention engine 1632, a multi-head attention engine 1634 over the output of the encoder 1610, and a fully-connected feed-forward network 1626. Each layer includes a residual connection (not shown) around the layer, which is followed by layer normalization. The masked multi -head self-attention engine 1632 is masked to prevent positions from attending to subsequent positions and ensures that the predictions at position i can depend only on the known outputs at positions less than i (e.g., auto-regression).

[0232] In the transformer, the queries, keys, and values are linearly projected by a multi -head attention engine into learned linear projects, and then attention is performed in parallel on each of the learned linear projects, which are concatenated and then projected into final values.

[0233] The transformer also includes a positional encoder 1640 to encode positions because the model does not contain recurrence and convolution, and relative or absolute position of the tokens is needed. In the transformer 1600, the positional encodings are added to the input embeddings at the bottom layer of the encoder 1610 and the decoder 1630. The positional encodings are summed with the embeddings because the positional encodings and embeddings have the same dimensions. A corresponding position decoder 1650 is configured to decode the positions of the embeddings for the decoder 1630.

[0234] In some aspects, the transformer 1600 uses self-attention mechanisms to selectively weigh the importance of different parts of an input sequence during processing and allows the model to attend to different parts of the input sequence while generating the output. The input sequence is first embedded into vectors and then passed through multiple layers of self-attention and feed-forward networks. The transformer 1600 can process input sequences of variable length, making it well-suited for natural language processing tasks where input lengths can vary greatly. Additionally, the self-attention mechanism allows the transformer 1600 to capture long-range dependencies between words in the input sequence, which is difficult for RNNs and CNNs. The transformer with self-attention has achieved results in several natural language processing tasks that are beyond the capabilities of other neural networks and has become a popular choice for Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO61language and text applications. For example, the various large language models, such as a generative pretrained transformer (e.g., ChatGPT, etc.) and other current models are types of transformer networks.

[0235] FIG. 17 illustrates an example computing-device architecture 1700 of an example computing device which can implement the various techniques described herein. In some examples, the computing device can include a mobile device, a wearable device, an extended reality device (e.g., a virtual reality (VR) device, an augmented reality (AR) device, or a mixed reality (MR) device), a personal computer, a laptop computer, a video server, a vehicle (or computing device of a vehicle), or other device. For example, the computing-device architecture 1700 may include, implement, or be included in any or all of XR system 100 of FIG. 1, XR device 102 of FIG. 1, XR system 200 of FIG. 2, display device 204 of FIG. 2, processing device 206 of FIG. 2, companion device 208 of FIG. 2, XR system 300 of FIG. 3, display 304 of FIG. 3, XR system 400 of FIG. 4, XR device 504 of FIG. 5, system 600 of FIG. 6, system 700 of FIG. 7, system 800 of FIG. 8, system 900 of FIG. 9 and / or other devices, modules, or systems described herein. Additionally or alternatively, computing-device architecture 1700 may be configured to perform process 1100, and / or other process described herein.

[0236] The components of computing-device architecture 1700 are shown in electrical communication with each other using connection 1712, such as a bus. The example computing-device architecture 1700 includes a processing unit (CPU or processor) 1702 and computing device connection 1712 that couples various computing device components including computing device memory 1710, such as read only memory (ROM) 1708 and random-access memory (RAM) 1706, to processor 1702.

[0237] Computing-device architecture 1700 can include a cache of high-speed memory connected directly with, in close proximity7to, or integrated as part of processor 1702. Computing-device architecture 1700 can copy data from memoiy 1710 and / or the storage device 1714 to cache 1704 for quick access by processor 1702. In this way, the cache can provide a performance boost that avoids processor 1702 delays while waiting for data. These and other modules can control or be configured to control processor 1702 to perform various actions. Other computing device memory 1710 may be available for use as well. Memory 1710 can include multiple different types of memory with different performance characteristics. Processor 1702 can include any general-purpose processor Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO62and a hardware or software service, such as service 1 1716, service 2 1718, and service 3 1720 stored in storage device 1714, configured to control processor 1702 as well as a special-purpose processor where software instructions are incorporated into the processor design. Processor 1702 may be a self-contained system, containing multiple cores or processors, a bus, memory controller, cache, etc. A multi-core processor may be symmetric or asymmetric.

[0238] To enable user interaction with the computing-device architecture 1700, input device 1722 can represent any number of input mechanisms, such as a microphone for speech, a touch-sensitive screen for gesture or graphical input, keyboard, mouse, motion input, speech and so forth. Output device 1724 can also be one or more of a number of output mechanisms known to those of skill in the art, such as a display, projector, television, speaker device, etc. In some instances, multimodal computing devices can enable a user to provide multiple types of input to communicate with computing-device architecture 1700. Communication interface 1726 can generally govern and manage the user input and computing device output. There is no restriction on operating on any particular hardware arrangement and therefore the basic features here may easily be substituted for improved hardware or firmware arrangements as they are developed.

[0239] Storage device 1714 is a non-volatile memoiy and can be a hard disk or other types of computer readable media which can store data that are accessible by a computer, such as magnetic cassettes, flash memory cards, solid state memoiy devices, digital versatile discs (DVDs), cartridges, random-access memories (RAMs) 1706, read only memory (ROM) 1708, and hybrids thereof. Storage device 1714 can include services 1716, 1718, and 1720 for controlling processor 1702. Other hardware or software modules are contemplated. Storage device 1714 can be connected to the computing device connection 1712. In one aspect, a hardware module that performs a particular function can include the software component stored in a computer-readable medium in connection with the necessary hardware components, such as processor 1702, connection 1712, output device 1724, and so forth, to carry out the function.

[0240] The term “substantially,” in reference to a given parameter, property, or condition, may refer to a degree that one of ordinary skill in the art would understand that the given parameter, property, or condition is met with a small degree of variance, such as, for example, within acceptable manufacturing tolerances. By way of example.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO63depending on the particular parameter, property, or condition that is substantially met, the parameter, property, or condition may be at least 90% met, at least 95% met, or even at least 99% met.

[0241] Aspects of the present disclosure are applicable to any suitable electronic device (such as security systems, smartphones, tablets, laptop computers, vehicles, drones, or other devices) including or coupled to one or more active depth sensing systems. While described below with respect to a device having or coupled to one light projector, aspects of the present disclosure are applicable to devices having any number of light projectors and are therefore not limited to specific devices.

[0242] The term “device” is not limited to one or a specific number of physical objects (such as one smartphone, one controller, one processing system and so on). As used herein, a device may be any electronic device with one or more parts that may implement at least some portions of this disclosure. While the below description and examples use the term “device” to describe various aspects of this disclosure, the term “device” is not limited to a specific configuration, type, or number of objects. Additionally, the term “system” is not limited to multiple components or specific aspects. For example, a system may be implemented on one or more printed circuit boards or other substrates and may have movable or static components. While the below description and examples use the term “system” to describe various aspects of this disclosure, the term “system” is not limited to a specific configuration, type, or number of objects.

[0243] Specific details are provided in the description above to provide a thorough understanding of the aspects and examples provided herein. However, it will be understood by one of ordinary skill in the art that the aspects may be practiced without these specific details. For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including functional blocks including devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software. Additional components may be used other than those shown in the figures and / or described herein. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the aspects in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the aspects.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO64

[0244] Individual aspects may be described above as a process or method which is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be re-arranged. A process is terminated when its operations are completed but could have additional steps not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination can correspond to a return of the function to the calling function or the main function.

[0245] Processes and methods according to the above-described examples can be implemented using computer-executable instructions that are stored or otherwise available from computer-readable media. Such instructions can include, for example, instructions and data which cause or otherwise configure a general-purpose computer, special purpose computer, or a processing device to perform a certain function or group of functions. Portions of computer resources used can be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, firmware, source code, etc.

[0246] The term “computer-readable medium” includes, but is not limited to, portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing, containing, or carrying instruction(s) and / or data. A computer-readable medium may include a non-transitory medium in which data can be stored and that does not include carrier waves and / or transitory electronic signals propagating wirelessly or over wired connections. Examples of a non-transitory medium may include, but are not limited to, a magnetic disk or tape, optical storage media such as compact disk (CD) or digital versatile disk (DVD), flash memory, magnetic or optical disks, USB devices provided with non-volatile memory, networked storage devices, any suitable combination thereof, among others. A computer-readable medium may have stored thereon code and / or machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and / or receiving information, data, arguments, parameters, or memory7contents. Information, arguments,Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO65parameters, data, etc. may be passed, forwarded, or transmitted via any suitable means including memory sharing, message passing, token passing, network transmission, or the like.

[0247] In some aspects the computer-readable storage devices, mediums, and memories can include a cable or wireless signal containing a bit stream and the like. However, when mentioned, non-transitory computer-readable storage media expressly exclude media such as energy, carrier signals, electromagnetic waves, and signals per se.

[0248] Devices implementing processes and methods according to these disclosures can include hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof, and can take any of a variety of form factors. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks (e.g., a computer-program product) may be stored in a computer-readable or machine-readable medium. A processor(s) may perform the necessary tasks. Typical examples of form factors include laptops, smart phones, mobile phones, tablet devices or other small form factor personal computers, personal digital assistants, rackmount devices, standalone devices, and so on. Functionality described herein also can be embodied in peripherals or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example.

[0249] The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are example means for providing the functions described in the disclosure.

[0250] In the foregoing description, aspects of the application are described with reference to specific aspects thereof, but those skilled in the art will recognize that the application is not limited thereto. Thus, while illustrative aspects of the application have been described in detail herein, it is to be understood that the inventive concepts may be otherwise variously embodied and employed, and that the appended claims are intended to be construed to include such variations, except as limited by the prior art. Various features and aspects of the above-described application may be used individually or jointly. Further, aspects can be utilized in any number of environments and applications beyond those described herein without departing from the broader spirit and scope of the specification. The specification and drawings are, accordingly, to be regarded as Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO66illustrative rather than restrictive. For the purposes of illustration, methods were described in a particular order. It should be appreciated that in alternate aspects, the methods may be performed in a different order than that described.

[0251] One of ordinary skill will appreciate that the less than (“<”) and greater than (“>”) symbols or terminology' used herein can be replaced with less than or equal to (“<”) and greater than or equal to (“>”) symbols, respectively, without departing from the scope of this description.

[0252] Where components are described as being “configured to” perform certain operations, such configuration can be accomplished, for example, by designing electronic circuits or other hardware to perform the operation, by programming programmable electronic circuits (e.g., microprocessors, or other suitable electronic circuits) to perform the operation, or any combination thereof.

[0253] The phrase “coupled to” refers to any component that is physically connected to another component either directly or indirectly, and / or any component that is in communication with another component (e.g., connected to the other component over a wired or wireless connection, and / or other suitable communication interface) either directly or indirectly.

[0254] Claim language or other language reciting “at least one of’ a set and / or “one or more” of a set indicates that one member of the set or multiple members of the set (in any combination) satisfy the claim. For example, claim language reciting “at least one of A and B” or “at least one of A or B” means A, B, or A and B. In another example, claim language reciting “at least one of A, B, and C” or “at least one of A, B, or C” means A, B, C, or A and B, or A and C, or B and C, A and B and C, or any duplicate information or data (e.g., A and A. B and B. C and C. A and A and B, and so on), or any other ordering, duplication, or combination of A, B, and C. The language “at least one of’ a set and / or “one or more” of a set does not limit the set to the items listed in the set. For example, claim language reciting “at least one of A and B” or “at least one of A or B” may mean A, B, or A and B, and may additionally include items not listed in the set of A and B. The phrases “at least one” and “one or more” are used interchangeably herein.

[0255] Claim language or other language reciting “at least one processor configured to,” “at least one processor being configured to,” “one or more processors configured to,”Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO67■‘one or more processors being configured to,’’ or the like indicates that one processor or multiple processors (in any combination) can perform the associated operation(s). For example, claim language reciting “at least one processor configured to: X, Y, and Z” means a single processor can be used to perform operations X, Y, and Z; or that multiple processors are each tasked with a certain subset of operations X, Y, and Z such that together the multiple processors perform X, Y, and Z; or that a group of multiple processors work together to perform operations X, Y, and Z. In another example, claim language reciting “at least one processor configured to: X, Y, and Z’’ can mean that any single processor may only perform at least a subset of operations X, Y, and Z.

[0256] Where reference is made to one or more elements performing functions (e.g., steps of a method), one element may perform all functions, or more than one element may collectively perform the functions. When more than one element collectively performs the functions, each function need not be performed by each of those elements (e.g., different functions may be performed by different elements) and / or each function need not be performed in whole by only one element (e.g., different elements may perform different sub-functions of a function). Similarly, where reference is made to one or more elements configured to cause another element (e.g., an apparatus) to perform functions, one element may be configured to cause the other element to perform all functions, or more than one element may collectively be configured to cause the other element to perform the functions.

[0257] Where reference is made to an entity (e.g., any entity or device described herein) performing functions or being configured to perform functions (e.g., steps of a method), the entity may be configured to cause one or more elements (individually or collectively) to perform the functions. The one or more components of the entity may include at least one memory, at least one processor, at least one communication interface, another component configured to perform one or more (or all) of the functions, and / or any combination thereof. Where reference to the entity performing functions, the entity may be configured to cause one component to perform all functions, or to cause more than one component to collectively perform the functions. When the entity is configured to cause more than one component to collectively perform the functions, each function need not be performed by each of those components (e.g., different functions may be performed by different components) and / or each function need not be performed in whole by onlyPolsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO68one component (e.g., different components may perform different sub-functions of a function).

[0258] The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the aspects disclosed herein may be implemented as electronic hardware, computer software, firmware, or combinations thereof. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

[0259] The techniques described herein may also be implemented in electronic hardware, computer software, firmware, or any combination thereof. Such techniques may be implemented in any of a variety of devices such as general-purposes computers, wireless communication device handsets, or integrated circuit devices having multiple uses including application in wireless communication device handsets and other devices. Any features described as modules or components may be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. If implemented in software, the techniques may be realized at least in part by a computer-readable data storage medium including program code including instructions that, when executed, performs one or more of the methods described above. The computer-readable data storage medium may form part of a computer program product, which may include packaging materials. The computer-readable medium may include memory or data storage media, such as random-access memory (RAM) such as synchronous dynamic random-access memory (SDRAM), read-only memory (ROM), non-volatile randomaccess memory (NVRAM), electrically erasable programmable read-only memory (EEPROM), flash memory7, magnetic or optical data storage media, and the like. The techniques additionally, or alternatively, may be realized at least in part by a computer-readable communication medium that carries or communicates program code in the form of instructions or data structures and that can be accessed, read, and / or executed by a computer, such as propagated signals or waves.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO69

[0260] The program code may be executed by a processor, which may include one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, an application specific integrated circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Such a processor may be configured to perform any of the techniques described in this disclosure. A general-purpose processor may be a microprocessor; but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, such as, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. Accordingly, the term “processor,” as used herein may refer to any of the foregoing structure, any combination of the foregoing structure, or any other structure or apparatus suitable for implementation of the techniques described herein.

[0261] Illustrative aspects of the disclosure include:

[0262] Aspect 1. An apparatus for generating answers to questions, the apparatus comprising: at least one memory’; and at least one processor coupled to the at least one memory and configured to: detect obj ects in a scene of a user; obtain scene data associated with the scene; generate scene information based on the detected objects and the scene data; identify a region of interest (ROI) of an image of the scene; generate user information based on a question and user data associated with the user; process the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identify a video from among a plurality of videos that is relevant to the answer; and provide at least one of the answer or the video to the user.

[0263] Aspect 2. The apparatus of aspect 1, wherein the at least one processor is configured to: detect at least one of a quick-response (QR) code, a name associated with the scene, or a recognizable landmark in the image of the scene; identify the scene based on at least one of the quick-response (QR) code, the name associated with the scene, or the recognizable landmark; and identify a data storage associated w ith the scene based on identification of the scene, wherein the scene data is obtained from the data storage associated with the scene.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO70

[0264] Aspect 3. The apparatus of any one of aspects 1 or 2, wherein the scene information is generated by processing indications of the detected objects and the scene data using a retrieval-augmented-generation (RAG) model.

[0265] Aspect 4. The apparatus of any one of aspects 1 to 3, wherein the ROI is based on at least one of: a gaze of the user; a center of the image; text detected in the image; or objects detected in the image.

[0266] Aspect 5. The apparatus of any one of aspects 1 to 4, wherein the image is obtained by a scene-facing camera of a head-mounted device.

[0267] Aspect 6. The apparatus of any one of aspects 1 to 5. wherein the user information is generated by processing the question and the user data using a retrieval-augmented-generation (RAG) model.

[0268] Aspect 7. The apparatus of any one of aspects 1 to 6, wherein, to obtain the question, the at least one processor is configured to: record the question as spoken by the user; and convert the spoken question into text.

[0269] Aspect 8. The apparatus of any one of aspects 1 to 7, wherein the user data is obtained from a data storage associated with the user.

[0270] Aspect 9. The apparatus of any one of aspects 1 to 8, wherein, to process the question, the image of the scene, the ROI, the scene information, and the user information using the machine-learning model, the at least one processor is configured to format the question, the image of the scene, the ROI, the scene information, and the user information as a prompt for the machine-learning model.

[0271] Aspect 10. The apparatus of any one of aspects 1 to 9, wherein the machinelearning model comprises a large multimodal machine-learning model.

[0272] Aspect 11. The apparatus of any one of aspects 1 to 10, wherein identifying the video from among the plurality of videos comprises processing the answer and at least one of titles or labels of the plurality of videos using a retrieval-augmented-generation (RAG) model.

[0273] Aspect 12. The apparatus of any one of aspects 1 to 11, wherein the at least one processor is configured to: detect negations in at least one of the question or the answer;Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO71and in response to detecting a negation in at least one of the question or the answer, determine to provide the answer to the user.

[0274] Aspect 13. The apparatus of any one of aspects 1 to 12, wherein the at least one processor is configured to: detect negations in at least one of the question or the answer; and in response to not detecting a negation in at least one of the question or the answer, determine to provide the video to the user.

[0275] Aspect 14. The apparatus of any one of aspects 1 to 13, wherein, to provide the answer to the user, the at least one processor is configured to vocalize the answer using a speaker of a head-mounted device.

[0276] Aspect 15. The apparatus of any one of aspects 1 to 14, wherein, to provide the video to the user, the at least one processor is configured to play the video using a display of a head-mounted device.

[0277] Aspect 16. A method for generating answers to questions, the method comprising: detecting objects in a scene of a user; obtaining scene data associated with the scene; generating scene information based on the detected objects and the scene data; identifying a region of interest (ROI) of an image of the scene; generating user information based on a question and user data associated with the user; processing the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer; identifying a video from among a plurality of videos that is relevant to the answer; and providing at least one of the answer or the video to the user.

[0278] Aspect 17. The method of aspect 16, further comprising: detecting at least one of a quick-response (QR) code, a name associated with the scene, or a recognizable landmark in the image of the scene; identifying the scene based on at least one of the quick-response (QR) code, the name associated with the scene, or the recognizable landmark; and identifying a data storage associated with the scene based on identification of the scene, wherein the scene data is obtained from the data storage associated with the scene.

[0279] Aspect 18. The method of any one of aspects 16 or 17, wherein the scene information is generated by processing indications of the detected objects and the scene data using a retrieval-augmented-generation (RAG) model.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO72

[0280] Aspect 19. The method of any one of aspects 16 to 18, wherein the ROI is based on at least one of: a gaze of the user; a center of the image; text detected in the image; or objects detected in the image.

[0281] Aspect 20. The method of any one of aspects 16 to 19, wherein the image is obtained by a scene-facing camera of a head-mounted device.

[0282] Aspect 21. The method of any one of aspects 16 to 20, wherein the user information is generated by processing the question and the user data using a retrieval-augmented-generation (RAG) model.

[0283] Aspect 22. The method of any one of aspects 16 to 21. wherein obtaining the question comprises: recording the question as spoken by the user; and converting the spoken question into text.

[0284] Aspect 23. The method of any one of aspects 16 to 22, wherein the user data is obtained from a data storage associated with the user.

[0285] Aspect 24. The method of any one of aspects 16 to 23, wherein processing the question, the image of the scene, the ROI, the scene information, and the user information using the machine-learning model comprises formatting the question, the image of the scene, the ROI, the scene information, and the user information as a prompt for the machine-learning model.

[0286] Aspect 25. The method of any one of aspects 16 to 24, wherein the machinelearning model comprises a large multimodal machine-learning model.

[0287] Aspect 26. The method of any one of aspects 16 to 25, wherein identifying the video from among the plurality of videos comprises processing the answer and at least one of titles or labels of the plurality of videos using a retrieval-augmented-generation (RAG) model.

[0288] Aspect 27. The method of any one of aspects 16 to 26, further comprising: detecting negations in at least one of the question or the answer; and in response to detecting a negation in at least one of the question or the answer, determining to provide the answer to the user.

[0289] Aspect 28. The method of any one of aspects 16 to 27, further comprising: detecting negations in at least one of the question or the answer; and in response to not Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO73detecting a negation in at least one of the question or the answer, determining to provide the video to the user.

[0290] Aspect 29. The method of any one of aspects 16 to 28, wherein providing the answer to the user comprises vocalizing the answer using a speaker of a head-mounted device.

[0291] Aspect 30. The method of any one of aspects 16 to 29, wherein providing the video to the user comprises playing the video using a display of ahead-mounted device.

[0292] Aspect 31. A non-transitory computer-readable storage medium having stored thereon instructions that, when executed by at least one processor, cause the at least one processor to perform operations according to any of aspects 16 to 30.

[0293] Aspect 32. An apparatus, the apparatus comprising one or more means for performing operations according to any of aspects 16 to 30.Polsinelli Ref. No. 094922-856802

Claims

Qualcomm Ref. No. 2502633 WO74CLAIMS WHAT IS CLAIMED IS:

1. An apparatus for generating answers to questions, the apparatus comprising:at least one memory; andat least one processor coupled to the at least one memory7and configured to: detect an obj ect in a scene of a user;obtain scene data associated with the scene;generate scene information based on the detected object and the scene data;identify a region of interest (ROI) of an image of the scene; generate user information based on a question and user data associated with the user;process the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer;identify a video from among a plurality7of videos that is relevant to the answer; andprovide at least one of the ansyver or the video to the user.

2. The apparatus of claim 1. wherein the at least one processor is configured to:detect at least one of a quick-response (QR) code, a name associated with the scene, or a recognizable landmark in the image of the scene;identify the scene based on at least one of the quick-response (QR) code, the name associated with the scene, or the recognizable landmark; andidentify a data storage associated with the scene based on identification of the scene,wherein the scene data is obtained from the data storage associated with the scene.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO753. The apparatus of claim 1, wherein the scene information is generated by processing indications of the detected object and the scene data using a retrieval-augmented-generation (RAG) model.

4. The apparatus of claim 1, wherein the ROI is based on at least one of a gaze of the user;a center of the image;text detected in the image; orobjects detected in the image.

5. The apparatus of claim 1, wherein the image is obtained by a scenefacing camera of a head-mounted device.

6. The apparatus of claim 1, wherein the user information is generated by processing the question and the user data using a retrieval-augmented-generation (RAG) model.

7. The apparatus of claim 1, wherein, to obtain the question, the at least one processor is configured to:record the question as spoken by the user; andconvert the recorded question into text.

8. The apparatus of claim 1, wherein the user data is obtained from a data storage associated with the user.

9. The apparatus of claim 1. wherein, to process the question, the image of the scene, the ROI, the scene information, and the user information using the machinelearning model, the at least one processor is configured to format the question, the image of the scene, the ROI, the scene information, and the user information as a prompt for the machine-learning model.

10. The apparatus of claim 1, wherein the machine-learning model comprises a large multimodal machine-learning model.Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO7611. The apparatus of claim 1, wherein identifying the video from among the plurality of videos comprises processing the answer and at least one of titles or labels of the plurality of videos using a retrieval-augmented-generation (RAG) model.

12. The apparatus of claim 1, wherein the at least one processor is configured to:detect negations in at least one of the question or the answer; andin response to detecting a negation in at least one of the question or the answer, determine to provide the answer to the user.

13. The apparatus of claim 1, wherein the at least one processor is configured to:detect negations in at least one of the question or the answer; andin response to not detecting a negation in at least one of the question or the answer, determine to provide the video to the user.

14. The apparatus of claim 1, wherein, to provide the answer to the user, the at least one processor is configured to vocalize the answer using a speaker of a headmounted device.

15. The apparatus of claim 1. wherein, to provide the video to the user, the at least one processor is configured to play the video using a display of a head-mounted device.

16. A method for generating answers to questions, the method comprising: detecting an object in a scene of a user;obtaining scene data associated with the scene;generating scene information based on the detected object and the scene data; identifying a region of interest (ROI) of an image of the scene;generating user information based on a question and user data associated with the user;Polsinelli Ref. No. 094922-856802Qualcomm Ref. No. 2502633 WO77processing the question, the image of the scene, the ROI, the scene information, and the user information using a machine-learning model to generate an answer;identifying a video from among a plurality of videos that is relevant to the answer; andproviding at least one of the answer or the video to the user.

17. The method of claim 16, further comprising:detecting at least one of a quick-response (QR) code, a name associated with the scene, or a recognizable landmark in the image of the scene;identifying the scene based on at least one of the quick -response (QR) code, the name associated with the scene, or the recognizable landmark; andidentifying a data storage associated with the scene based on identification of the scene,wherein the scene data is obtained from the data storage associated with the scene.

18. The method of claim 16, wherein the scene information is generated by processing indications of the detected object and the scene data using a retrieval-augmented-generation (RAG) model.

19. The method of claim 16, wherein the ROI is based on at least one of: a gaze of the user;a center of the image;text detected in the image; orobjects detected in the image.

20. The method of claim 16, wherein the image is obtained by a scene-facing camera of a head-mounted device.Polsinelli Ref. No. 094922-856802

Images