Attention-based refinement for depth completion

Abstract

A processor-implemented method for attention-based deep completion includes receiving an input through an artificial neural network (ANN). The input includes an image and a sparse depth measurement. The ANN extracts the input multi-scale visual features. The ANN applies a self-attention mechanism to the multi-scale visual features to generate a set of attention processed multi-scale visual features. The ANN generates a dense depth map based on the set of attention-processed multi-scale visual features.

Description

Cross-references to related applications

[0001] This application claims priority to U.S. Patent Application No. 18 / 448,845, filed August 11, 2023, entitled “ATTENTION-BASED REFINEMENT FORDEPTH COMPLETION”, the entire disclosure of which is expressly incorporated herein by reference. Technical Field

[0002] The various aspects of this disclosure relate to attention-based refinement for deep completion. Background Technology

[0003] Artificial neural networks can comprise interconnected groups of artificial neurons (e.g., neuron models). Artificial neural networks can be computing devices or represented as methods to be performed by computing devices. Convolutional neural networks (CNNs) are a type of feedforward artificial neural network. A CNN can comprise an ensemble of neurons, where each neuron has a receptive field and collectively constructs the input space. CNNs, such as deep convolutional neural networks (DCNs), have numerous applications. Specifically, these neural network architectures are used in various technologies such as image recognition, speech recognition, acoustic scene classification, keyword retrieval, autonomous driving, and other classification tasks.

[0004] Depth completion involves estimating dense, pixel-by-pixel depth from a sparse map captured by a depth sensor. It is an important task in fields such as autonomous driving, robotics, simultaneous localization and mapping (SMR), extended reality (XR), and 3D perception. Such applications are expected to be utilized on edge devices such as smartphones. However, conventional depth completion methods rely on iterative processing, which can be computationally complex, involving prohibitive memory and power budgets, and in some cases, resource-constrained devices may not support such processing. Summary of the Invention

[0005] In some aspects of this disclosure, a processor-implemented method executed by at least one processor includes: receiving input via an artificial neural network (ANN), the input including an image and sparse depth measurements. The method further includes extracting multi-scale visual features from the input via the ANN. The method further includes applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features. The method further includes generating a dense depth map based on the set of attention-processed multi-scale visual features via the ANN.

[0006] Various aspects of this disclosure relate to an apparatus comprising: means for receiving input via an artificial neural network (ANN), the input including an image and sparse depth measurements; means for extracting multi-scale visual features of the input via the ANN; means for applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features; and means for generating a dense depth map via the ANN based on the set of attention-processed multi-scale visual features.

[0007] In some aspects of this disclosure, a non-transitory computer-readable medium having non-transitory program code recorded thereon is disclosed. The program code is executed by a processor and includes: program code for receiving input via an artificial neural network (ANN), the input including an image and sparse depth measurements; program code for extracting multi-scale visual features of the input via the ANN; program code for applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features; and program code for generating a dense depth map via the ANN based on the set of attention-processed multi-scale visual features.

[0008] Various aspects of this disclosure relate to an apparatus having a memory and one or more processors coupled to the memory. The processors are configured to receive input via an artificial neural network (ANN), the input including an image and sparse depth measurements. The processors are also configured to extract multi-scale visual features from the input via the ANN. The processors are further configured to apply a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features. The processors are further configured to generate a dense depth map based on the set of attention-processed multi-scale visual features via the ANN.

[0009] Additional features and advantages of this disclosure will be described below. Those skilled in the art will understand that this disclosure can be readily used as the basis for modifying or designing other structures for implementing the same purposes as this disclosure. Those skilled in the art will also recognize that such equivalent constructions do not depart from the teachings of this disclosure as set forth in the appended claims. Novel features considered characteristic of this disclosure, in both their organization and manner of operation, along with further objects and advantages, will be better understood when considered in conjunction with the accompanying drawings. However, it is to be clearly understood that each drawing is provided for illustrative and descriptive purposes only and is not intended to be a definition of a limitation of this disclosure. Attached Figure Description

[0010] The features, substance, and advantages of this disclosure will become more apparent when understood in conjunction with the accompanying drawings, in which the same reference numerals are consistently used for identification.

[0011] Figure 1 Example implementations of neural networks using a system-on-a-chip (SoC) (including a general-purpose processor) according to certain aspects of this disclosure are illustrated.

[0012] Figure 2A , Figure 2B and Figure 2C These are illustrations of neural networks according to various aspects of this disclosure.

[0013] Figure 2D This is a diagram illustrating exemplary deep convolutional networks (DCNs) according to various aspects of this disclosure.

[0014] Figure 3 This is a block diagram illustrating exemplary deep convolutional networks (DCNs) according to various aspects of this disclosure.

[0015] Figure 4 This is a block diagram illustrating an exemplary software architecture for modular artificial intelligence (AI) functions according to various aspects of this disclosure.

[0016] Figure 5 This is a flowchart illustrating a conventional depth completion method.

[0017] Figure 6 This is a block diagram illustrating an example architecture for image-guided depth completion according to various aspects of this disclosure.

[0018] Figure 7 This is a flowchart illustrating a method for an attention-based refinement processor implementation for deep completion according to various aspects of this disclosure. Detailed Implementation

[0019] The detailed description following, taken in conjunction with the accompanying drawings, is intended as a description of various configurations and not as representing only configurations in which the described concepts can be practiced. To provide a comprehensive understanding of the various concepts, the detailed description includes specific details. However, it will be apparent to those skilled in the art that these concepts can be practiced without these specific details. In some instances, to avoid obscuring such concepts, well-known structures and components are shown in block diagram form.

[0020] Based on the teachings, those skilled in the art will recognize that the scope of this disclosure is intended to cover any aspect of this disclosure, whether implemented independently of or in combination with any other aspect of this disclosure. For example, an apparatus or method may be implemented using any number of the aspects described. Furthermore, the scope of this disclosure is intended to cover such apparatuses or methods practiced using other structures, functionalities, or structures and functionalities that complement or differ from the various aspects of this disclosure described. It should be understood that any aspect of this disclosure may be embodied by one or more elements of the claims.

[0021] The word “exemplary” is used to mean “serving as an example, instance, or illustration.” Any aspect described as “exemplary” need not be interpreted as superior to or better than other aspects.

[0022] While specific aspects have been described, numerous variations and substitutions of these aspects fall within the scope of this disclosure. Although some benefits and advantages of preferred aspects have been mentioned, the scope of this disclosure is not intended to be limited to a particular benefit, use, or purpose. Rather, aspects of this disclosure are intended to be broadly applicable to different technologies, system configurations, networks, and protocols, some of which are illustrated by way of example in the accompanying drawings and the following description of preferred aspects. The detailed description and drawings are merely illustrative and not limiting of this disclosure, the scope of which is defined by the appended claims and their equivalents.

[0023] As described above, depth completion is an important task in fields such as autonomous driving, robotics, simultaneous localization and mapping (SLAM), extended reality (XR), and 3D perception. Depth completion involves estimating dense, pixel-by-pixel depth from a sparse map captured by a depth sensor. In other words, given an image captured by a single camera and sparse depth measurements, the distance of each pixel from the camera can be determined.

[0024] Conventional methods rely on iterative processing and thinning for depth completion. However, iterative processing utilizes deformable convolution or dynamic graph propagation. Deformable convolution adds a two-dimensional offset to the regular grid sampling locations in standard convolution operations, deforming the receptive field during input sampling. Deformable convolution allows the network to capture finer-grained spatial patterns of objects in an image. On the other hand, dynamic graph propagation updates the initial depth value of a pixel based on its neighborhood and an affinity matrix that defines the amount of information passed between neighboring pixels.

[0025] Deformable convolutions and dynamic graph propagation can both produce high-quality depth completion. However, such methods are computationally complex and can consume significant computational and storage resources. Additionally, on-device operations may not support these methods. Therefore, performing depth completion on edge devices is challenging.

[0026] Therefore, to address these and other challenges, aspects of this disclosure relate to attention-based refinement for depth completion. According to various aspects of this disclosure, an artificial neural network (ANN) can process inputs including images and sparse depth measurements to extract multi-scale visual features. The ANN can apply self-attention to the extracted multi-scale visual features and generate a depth map based on the attention-processed multi-scale visual features.

[0027] Specific aspects of the subject matter described in this disclosure may be implemented to achieve one or more of the following potential advantages. In some examples, the described techniques (e.g., applying self-attention to extracted multi-scale visual features and generating a depth map based on the attention-processed multi-scale visual features) may reduce memory and / or power consumption, and / or enable depth completion to be performed on mobile devices.

[0028] Figure 1 An example implementation of a system-on-a-chip (SOC) 100 is illustrated, which may include a central processing unit (CPU) 102 or an attention-based, refined multi-core CPU configured for deep completion. Variables (e.g., neural signals and synaptic weights), system parameters associated with the computing device (e.g., a weighted neural network), latency, frequency bin information, and task information may be stored in a memory block associated with a neural processing unit (NPU) 108, a memory block associated with the CPU 102, a memory block associated with a graphics processing unit (GPU) 104, a memory block associated with a digital signal processor (DSP) 106, a memory block 118, or may be distributed across multiple blocks. Instructions executed at the CPU 102 may be loaded from the program memory associated with the CPU 102 or from memory block 118.

[0029] SOC 100 may also include additional processing blocks tailored for specific functions, such as GPU 104, DSP 106, connectivity block 110 (which may include fifth-generation (5G) connectivity, fourth-generation LTE (4G) connectivity, Wi-Fi connectivity, USB connectivity, Bluetooth connectivity, etc.), and multimedia processor 112 capable of, for example, detecting and recognizing gestures. In one specific implementation, NPU 108 is implemented within CPU 102, DSP 106, and / or GPU 104. SOC 100 may also include sensor processor 114, image signal processor (ISP) 116, and / or navigation module 120, which may include a global positioning system.

[0030] The SOC 100 may be based on the ARM instruction set. In one aspect of this disclosure, the instructions loaded into the general-purpose processor 102 may include code for receiving input by an artificial neural network (ANN). The input includes an image and sparse depth measurements. The general-purpose processor 102 may also include code for extracting multi-scale visual features from the input via the ANN. The general-purpose processor 102 may include code for applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features. The general-purpose processor 102 may also include code for generating a dense depth map via the ANN based on the set of attention-processed multi-scale visual features.

[0031] Deep learning architectures perform object recognition tasks by learning to represent inputs at progressively higher levels of abstraction in each layer, thereby constructing useful feature representations of the input data. In this way, deep learning addresses a major bottleneck of traditional machine learning. Before deep learning, machine learning methods for object recognition problems often relied heavily on human-designed features, possibly in conjunction with shallow classifiers. Shallow classifiers could be two-class linear classifiers, where a weighted sum of feature vector components is compared to a threshold to predict which class the input belongs to. Human-designed features could be templates or kernels customized for a specific problem domain by engineers with domain expertise. In contrast, while deep learning architectures can learn to represent features similar to those that human engineers might design, this requires training. Furthermore, deep networks can learn to represent and recognize novel types of features that humans might not have considered.

[0032] Deep learning architectures can learn hierarchical structures of features. For example, if presented with visual data, the first layer can learn to recognize relatively simple features in the input stream, such as edges. In another example, if presented with auditory data, the first layer can learn to recognize spectral power at specific frequencies. The second layer, taking the output of the first layer as input, can learn to recognize combinations of features, such as simple shapes in visual data or combinations of sounds in auditory data. For example, higher layers can learn to represent complex shapes in visual data or words in auditory data. Even higher layers can learn to recognize common visual objects or spoken phrases.

[0033] Deep learning architectures perform particularly well when applied to problems with a natural hierarchical structure. For example, the classification of motorized vehicles can benefit from first learning to identify features such as wheels, windshields, and others. These features can then be combined in different ways at higher levels to identify cars, trucks, and airplanes.

[0034] Neural networks can be designed to have multiple connectivity patterns. In feedforward networks, information is passed from lower layers to higher layers, where each neuron in a given layer communicates with neurons in higher layers. As described above, hierarchical representations can be built in successive layers of a feedforward network. Neural networks can also have recurrent or feedback (also known as top-down) connections. In recurrent connections, the output from a neuron in a given layer can be passed to another neuron in the same layer. Recurrent architectures can help identify patterns across more than one block of input data that is sequentially delivered to the neural network. Connections from neurons in a given layer to neurons in lower layers are called feedback (or top-down) connections. Networks with many feedback connections can be helpful when the recognition of higher-level concepts can aid in discerning specific lower-level features of the input.

[0035] The connections between layers in a neural network can be fully connected or locally connected. Figure 2A An example of a fully connected neural network 202 is illustrated. In the fully connected neural network 202, neurons in the first layer can transmit their outputs to each neuron in the second layer, so that each neuron in the second layer will receive inputs from each neuron in the first layer. Figure 2B An example of a locally connected neural network 204 is illustrated. In the locally connected neural network 204, neurons in the first layer can connect to a finite number of neurons in the second layer. More generally, the locally connected layers of the locally connected neural network 204 can be configured such that each neuron in the layer will have the same or similar connectivity pattern, but the connection strength can have different values ​​(e.g., 210, 212, 214, and 216). The connectivity pattern of locally connected layers can produce spatially different receptive fields in higher layers because higher-layer neurons in a given region can receive inputs that are tuned to the characteristics of a restricted portion of the total input to the network through training.

[0036] An example of a locally connected neural network is a convolutional neural network. Figure 2C An example of a convolutional neural network 206 is illustrated. Convolutional neural network 206 can be configured such that the connection strength associated with the input for each neuron in the second layer is shared (e.g., 208). Convolutional neural networks may be well-suited for problems where the spatial location of the input is meaningful.

[0037] One type of convolutional neural network is the deep convolutional network (DCN). Figure 2DA detailed example of a DCN 200 designed to recognize visual features from an image 226 input from an image capture device 230 (such as an in-vehicle camera) is illustrated. The DCN 200 in this example can be trained to identify traffic signs and the numbers provided on them. Of course, the DCN 200 can be trained for other tasks, such as identifying lane markings or traffic lights.

[0038] Supervised learning can be used to train the DCN 200. During training, an image (such as image 226 of a speed limit sign) can be presented to the DCN 200, and forward passes can then be computed to produce output 222. The DCN 200 may include a feature extraction part and a classification part. Upon receiving image 226, convolutional layer 232 can apply a convolutional kernel (not shown) to image 226 to generate a first set 218 of feature maps. As an example, the convolutional kernel used for convolutional layer 232 may be a 5x5 kernel that generates a 28x28 feature map. In this example, because four different feature maps are generated in the first set of feature maps 218, four different convolutional kernels are applied to image 226 at convolutional layer 232. Convolutional kernels may also be referred to as filters or convolutional filters.

[0039] The first set of feature maps 218 can be subsampled by a max-pooling layer (not shown) to generate a second set of feature maps 220. The max-pooling layer reduces the size of the first set of feature maps 218. That is, the size of the second set of feature maps 220 (e.g., 14x14) is smaller than the size of the first set of feature maps 218 (e.g., 28x28). The reduced size provides similar information to subsequent layers while reducing memory consumption. The second set of feature maps 220 can be further convolved via one or more subsequent convolutional layers (not shown) to generate one or more subsequent sets of feature maps (not shown).

[0040] exist Figure 2D In the example, the second set of feature maps 220 is convolved to generate a first feature vector 224. Furthermore, the first feature vector 224 is further convolved to generate a second feature vector 228. Each feature of the second feature vector 228 may include a number corresponding to a possible feature of image 226, such as "sign", "60", and "100". A softmax function (not shown) can convert the numbers in the second feature vector 228 into probabilities. Thus, the output 222 of DCN 200 can be the probability that image 226 includes one or more features.

[0041] In this example, the probabilities for "sign" and "60" in output 222 are higher than the probabilities for other numbers in output 222 (such as "30", "40", "50", "70", "80", "90", and "100"). Before training, output 222 generated by DCN 200 may be incorrect. Therefore, the error between output 222 and the target output can be calculated. The target output is the ground truth of image 226 (e.g., "sign" and "60"). The weights of DCN 200 can then be adjusted so that output 222 of DCN 200 is more closely aligned with the target output.

[0042] To adjust the weights, the learning algorithm computes the gradient vector of the weights. The gradient indicates by how much the error will increase or decrease as the weights are adjusted. At the top layers, the gradient corresponds directly to the values ​​of the weights connecting the activated neurons in the penultimate layer to the neurons in the output layer. In lower layers, the gradient depends on the values ​​of the weights and the error gradient computed in the higher layers. The weights can then be adjusted to reduce the error. This method of adjusting weights is called "backpropagation" because it involves "passing backward" through the neural network.

[0043] In practice, the error gradient of the weights can be calculated using a small number of examples to make the calculated gradient approximate the true error gradient. This approximation method is called stochastic gradient descent. Stochastic gradient descent can be repeated until the achievable error rate of the entire system stops decreasing or until the error rate reaches a target level. After learning, a new image (e.g., a speed limit sign in image 226) can be presented to DCN 200, and output 222 can be generated through the forward pass of DCN 200. This output can be considered as an inference or prediction of DCN 200.

[0044] Deep Belief Networks (DBNs) are probabilistic models that include multiple layers of hidden nodes. DBNs can be used to extract hierarchical representations of training datasets. DBNs are obtained by stacking layers of Restricted Boltzmann Machines (RBMs). An RBM is a type of artificial neural network that learns a probability distribution from a set of inputs. Because RBMs can learn a probability distribution without information about the class each input should be classified into, they are often used for unsupervised learning. Using a hybrid paradigm of supervised and unsupervised learning, the bottom RBM of a DBN can be trained unsupervised and used as a feature extractor, while the top RBM can be trained supervisedly (on the joint distribution of inputs from the previous layer and the target class) and used as a classifier.

[0045] DCN is a network of convolutional networks configured with additional pooling and normalization layers. DCN has achieved state-of-the-art performance on many tasks. DCN can be trained using supervised learning, where both the input and output targets are known for many paradigms and are used to modify the network's weights using gradient descent.

[0046] DCNs can be feedforward networks. Furthermore, as described above, connections from neurons in the first layer of a DCN to a set of neurons in the next higher layer are shared across neurons in the first layer. The feedforward and shared connections of a DCN can be used for fast processing. For example, the computational cost of a DCN may be much smaller than that of a similarly sized neural network that includes recurrent or feedback connections.

[0047] The processing at each layer of a convolutional network can be thought of as a spatially invariant template or base projection. If the input is first decomposed into multiple channels, such as the red, green, and blue channels of a color image, then a convolutional network trained on that input can be thought of as three-dimensional, with two spatial dimensions along the axes of the image and a third dimension capturing color information. The output of the convolutional connections can be thought of as forming a feature map in the next layer, where each element in the feature map (e.g., 220) receives input from a range of neurons in the previous layer (e.g., feature map 218) and from each of the multiple channels. The values ​​in the feature map can be further processed with nonlinearities (such as correction, max(0,x)). Values ​​from neighboring neurons can be further pooled, which corresponds to downsampling and provides additional local invariance and dimensionality reduction. Normalization corresponding to whitening can also be applied through lateral inhibition between neurons in the feature map.

[0048] Figure 3 This is a block diagram illustrating a DCN 350. A DCN 350 can include multiple layers of different types based on connectivity and weight sharing. For example... Figure 3 As shown, DCN 350 includes convolutional blocks 354A and 354B. Each convolutional block in convolutional blocks 354A and 354B can be configured with a convolutional layer (CONV) 356, a normalization layer (LNorm) 358, and a max pooling layer (MAX POOL) 360.

[0049] Although only two of the convolutional blocks 354A and 354B are shown, this disclosure is not limited thereto, and any number of convolutional blocks 354A and 354B may be included in the DCN 350 according to design preferences.

[0050] Convolutional layer 356 may include one or more convolutional filters that can be applied to the input data to generate feature maps. Normalization layer 358 can normalize the output of the convolutional filters. For example, normalization layer 358 can provide whitening or lateral suppression. Max pooling layer 360 can provide spatial downsampling aggregation to achieve local invariance and dimensionality reduction.

[0051] Parallel filter banks of deep convolutional networks can be loaded onto an SOC 100 (e.g., Figure 1 The CPU 102 or GPU 104 of the SOC 100 can be used to achieve high performance and low power consumption. In an alternative implementation, a parallel filter bank can be loaded onto the DSP 106 or ISP 116 of the SOC 100. Additionally, the DCN 350 can access other processing blocks that may exist on the SOC 100, such as the sensor processor 114 and navigation module 120, which are dedicated to sensors and navigation, respectively.

[0052] The DCN 350 may also include one or more fully connected layers 362 (FC1 and FC2). The DCN 350 may also include logistic regression (LR) layers 364. Weights (not shown) to be updated are located between each of the layers 356, 358, 360, 362, and 364 of the DCN 350. The output of each layer (e.g., 356, 358, 360, 362, and 364) can be used as input to the next layer in the DCN 350 (e.g., 356, 358, 360, 362, and 364) to learn hierarchical feature representations from the input data 352 (e.g., images, audio, video, sensor data, and / or other input data) supplied at the first convolutional block in convolutional block 354A. The output of the DCN 350 is a classification score 366 of the input data 352. The classification score 366 may be a set of probabilities, where each probability is the probability that the input data includes a feature from the feature set.

[0053] Figure 4 This is a block diagram illustrating an exemplary software architecture 400 with modular artificial intelligence (AI) capabilities. Using architecture 400, applications can be designed that enable SOC 420 (which can be similar to...) Figure 1 Various processing blocks of the SOC 100 (e.g., CPU 422, DSP 424, GPU 426, and / or NPU 428) support attention-based refinement for deep completion of AI applications 402 according to various aspects of this disclosure. Architecture 400 may, for example, be included in a computing device such as a smartphone.

[0054] AI application 402 can be configured to invoke functions defined in user space 404, which may, for example, provide the detection and recognition of a scene indicating the current location of the computing device (including architecture 400). For example, AI application 402 may configure microphones and cameras differently depending on whether the recognized scene is an office, lecture hall, restaurant, or an outdoor environment such as a lake. AI application 402 may make requests to compiled program code associated with libraries defined in the AI ​​Function Application Programming Interface (API) 406. This request may ultimately rely on the output of a deep neural network configured to provide inferred responses based on, for example, video and location data.

[0055] Runtime engine 408 (which may be compiled code of a runtime framework) may be further accessible to AI application 402. AI application 402 may cause runtime engine 408 to request inference, for example, at specific time intervals or when triggered by events detected by the user interface of AI application 402. When runtime engine 408 provides an inference response, it may then signal to the operating system (OS) space 410 running on SOC 420, such as kernel 412. In some examples, kernel 412 may be a LINUX kernel. The operating system may then enable sequential relaxation of quantization to be performed on CPU 422, DSP 424, GPU 426, NPU 428, or some combination thereof. CPU 422 may be directly accessible by the operating system, while other processing blocks may be accessed via drivers such as drivers 414, 416, or 418 for DSP 424, GPU 426, or NPU 428, respectively. In an exemplary example, the deep neural network may be configured to run on a combination of processing blocks such as CPU 422, DSP 424 and GPU 426, or on NPU 428.

[0056] Depth completion involves estimating dense, pixel-by-pixel depth from a sparse map captured by a depth sensor. Given an image captured by a single camera and sparse depth measurements (e.g., from a light detection and ranging (LiDAR) or time-of-flight sensor), depth completion aims to estimate the distance of each pixel from the camera.

[0057] Deep completion plays an important role in a variety of applications, including extended reality (XR), autonomous driving, camera image or video processing, and robotics.

[0058] Conventional depth completion techniques rely on iterative processing or thinning to achieve high-quality depth completion. Furthermore, iterative processing involves deformable convolutions or dynamic graph propagation. These operations are very time-consuming and computationally expensive from a power and memory perspective, which can be exacerbated in resource-constrained devices such as smartphones or other mobile devices. Additionally, deformable convolutions, dynamic graph propagation, and dynamic spatial propagation may not be supported on the device itself.

[0059] Therefore, aspects of this disclosure relate to attention-based refinement for depth completion. According to aspects of this disclosure, an attention model can be used as a refinement module. The neural network model may include a base neural network, such as a convolutional neural network (CNN) or a sparse-to-dense (S2D) network, which can receive an image and sparse depth measurements. The image and sparse depth measurements can be stitched together and received as input by the base neural network. The base neural network can process the stitched input to extract multi-scale visual features (e.g., spatial or spectral features) of the stitched input. That is, visual features can be extracted at multiple different scales (e.g., resolution) of the image, which increases the range of features extracted.

[0060] Self-attention can be applied to extracted multi-scale visual features to enable feature interaction / aggregation. Self-attention can be thought of as a mechanism that compares each element of the input (e.g., multi-scale visual features) with other elements of the input (e.g., other multi-scale visual features) to identify parts of the input that require attention or focus. In other words, self-attention allows elements of the input to interact with each other. The interactions of each input can then be aggregated to determine an attention score for each element of the input. Elements with the highest attention scores (e.g., multi-scale visual features) can be considered to contribute more to the output (e.g., depth) and thus receive more attention. A neural network model can then generate a dense depth map based on the attention-processed multi-scale visual features.

[0061] Figure 5 This is block diagram 500 illustrating a conventional depth completion method. (Reference) Figure 5The conventional depth completion method 500 includes a residual network 502 (also known as a residual neural network or ResNet), a shared decoder 506, and a thinning module 508. The residual network 502 receives an image 510 and sparse depth measurements 512. The image 510 and sparse depth measurements 512 are concatenated and processed by the residual network 502. The residual network 502 extracts features from the concatenated image 510 and sparse depth measurements 512. The extracted features are then provided to the shared decoder 506, which decodes the extracted features to generate an initial depth estimate 514. The initial depth estimate 514 is then provided to the thinning module 508 to refine the initial depth estimate 514. As described above, the thinning module 508 implements an iterative process that involves repeatedly processing the initial depth estimate 514 through deformable convolution, dynamic graph propagation, or other spatial propagation techniques to generate a final depth estimate 516.

[0062] Figure 6 This is a block diagram illustrating an example architecture 600 for image-guided depth completion according to various aspects of this disclosure. References Figure 6 Example architecture 600 includes an artificial neural network (ANN) 602. The artificial neural network 602 may, for example, include a convolutional neural network (e.g., Figure 3 (e.g., 350). In some aspects, ANN 602 may include sparse to dense networks. Example architecture 600 may also include a self-attention module 604 and a decoder 606. In other aspects, ANN 602 may include any other suitable networks.

[0063] ANN 602 can receive input including image 510 and sparse depth measurement 512. In various aspects, image 510 may include a single image from a single camera. In some aspects, image 510 may include a two-dimensional (2D) image. Additionally, image 510 may include, for example, a red-green-blue (RGB) image. Although a single image 510 is shown, the ANN can be configured to receive and process image sequences, such as video or streaming multimedia input. For example, video may include a sequence of video frames, where each frame of the video can be considered an image.

[0064] For example, sparse depth measurement 512 may include (but is not limited to) LiDAR measurement, RGB depth (RGBD) measurement, time-of-flight (ToF) measurement, or other types of sparse depth measurement. In some aspects, image 510 and sparse depth measurement 512 may be stitched together.

[0065] The input (e.g., 510, 512, or a concatenation of 510 and 512) can be fed to ANN 602. The ANN can extract multi-scale visual features from the input. The multi-scale visual features can be fed to the self-attention module 604.

[0066] In some respects, depthwise separable convolution can be applied to a subset of the scales of the input (e.g., four out of five scales). Depthwise separable convolution can split the convolutional kernel into multiple smaller kernels, each applied to a different channel (e.g., an RGB input could have three channels corresponding to a color). Depthwise convolution and pointwise convolution can then be performed on each channel of the input. Therefore, in some respects, depthwise convolution and pointwise convolution can be performed on a subset of multi-scale visual features. Furthermore, in some respects, the lowest scale of the multi-scale visual features can be downsampled.

[0067] The self-attention module 604 can apply self-attention to multi-scale visual features to generate attention-processed multi-scale visual features. In various aspects of this disclosure, self-attention can be applied to downsampled features of multi-scale visual features. In this case, depthwise separable deconvolution can be applied to the attention-processed multi-scale visual features to recover the feature size.

[0068] Attention-processed multi-scale visual features can be supplied to decoder 606. Decoder 606 can decode the attention-processed multi-scale visual features to generate dense depth map 608 (sometimes also called final depth estimate).

[0069] Advantageously, ANN 602 can generate dense depth maps 608 without employing the iterative refinement (e.g., refinement module 508) used in the conventional depth completion method 500. That is, the example architecture 600 can generate dense depth maps 608 by applying self-attention (e.g., using self-attention module 604) to the extracted multi-scale visual features instead of using deformable convolutions or dynamic graph propagation. As described above, iterative refinement, including deformable convolutions, dynamic graph propagation, or other spatial propagation techniques, is computationally complex and can involve significant memory and power consumption.

[0070] Furthermore, unlike conventional techniques involving deformable convolutions or dynamic graph propagation, example architecture 600 utilizes self-attention (e.g., module 604), which can be supported on the device side. Thus, using example architecture 600, dense depth graph 608 (e.g., depth completion) can be performed on edge devices such as smartphones or other mobile devices.

[0071] Figure 7 This is a flowchart illustrating a processor implementation of an attention-based refinement method 700 for deep completion according to various aspects of this disclosure. For example, the processor implementation of method 700 may be executed by one or more processors, such as CPUs (e.g., 102, 422), GPUs (e.g., 104, 426), DSPs (e.g., 106, 424), and / or NPUs (e.g., 108, 428)).

[0072] like Figure 7 As shown, at box 702, the at least one processor receives input via an artificial neural network (ANN). This input includes an image and sparse depth measurements. For example, as described with respect to various aspects, image 510 may include a single image from a single camera. In some aspects, image 510 may include, for example, a red-green-blue (RGB) image. For example, sparse depth measurement 512 may include LiDAR measurements, RGB depth (RGBD) measurements, time-of-flight (ToF) measurements, or other types of sparse depth measurements. In some aspects, image 510 and sparse depth measurement 512 may be stitched together.

[0073] At box 704, the at least one processor extracts multi-scale visual features of the input through the ANN. For example, as referenced... Figure 6 As described, ANN 602 can extract multi-scale visual features from the input. In some aspects, depthwise separable convolutions can be applied to a subset of the input scales (e.g., to four out of five scales). Furthermore, in some aspects, the lowest scale of the multi-scale visual features can be downsampled.

[0074] At box 706, the at least one processor applies a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features. For example, as referenced Figure 6 As described, the self-attention module 604 can apply self-attention to multi-scale visual features to generate attention-processed multi-scale visual features. In some aspects, self-attention can be applied to downsampled features of multi-scale visual features. In this case, depthwise separable deconvolution can be applied to the attention-processed multi-scale visual features to recover the feature size.

[0075] At box 708, the at least one processor generates a dense depth map using the ANN based on the set of attention-processed multi-scale visual features. For example, as referenced... Figure 6 As described, attention-processed multi-scale visual features can be supplied to decoder 606. Decoder 606 can decode the attention-processed multi-scale visual features to generate a dense depth map 608.

[0076] Specific implementation examples are provided in the following numbered clauses.

[0077] 1. A processor-implemented method executed by at least one processor, the processor-implemented method comprising: The input is received via an artificial neural network (ANN), which includes an image and sparse depth measurements; The ANN is used to extract multi-scale visual features from the input; The ANN is used to apply a self-attention mechanism to the multi-scale visual features to generate a set of attention-processed multi-scale visual features; and The ANN generates a dense depth map based on the attention-processed multi-scale visual features of the group.

[0078] 2. The processor-implemented method according to Clause 1, wherein the sparse depth measurement includes optical detection and ranging (LiDAR) measurement, red-green-blue depth (RGBD) measurement, or time-of-flight (ToF) measurement.

[0079] 3. The processor-implemented method according to Clause 1 or 2, wherein the processor-implemented method is executed by at least one processor of the mobile device.

[0080] 4. The processor-implemented method according to any one of Clauses 1 to 3, wherein the processor-implemented method further comprises implementing the dense depth map in an extended reality (XR) application, an autonomous driving application, a robotics application, or an image processing application.

[0081] 5. A processor-implemented method according to any one of clauses 1 to 4, the processor-implemented method further comprising processing the multi-scale visual features by applying depthwise separable convolutions to a subset of the multi-scale visual features via the ANN.

[0082] 6. A method implemented by a processor according to any one of Clauses 1 to 5, wherein the ANN comprises a sparse-to-dense (S2D) network.

[0083] 7. A method implemented by a processor according to any one of Clauses 1 to 6, wherein the ANN includes a convolutional neural network (CNN).

[0084] 8. A method implemented by a processor according to any one of Clauses 1 to 7, wherein the image is captured by a single camera.

[0085] 9. A processor-implemented method according to any one of Clauses 1 to 8, wherein the processor-implemented method is executed by at least one processor of a mobile device, wherein the single camera is included in the mobile device.

[0086] 10. An apparatus comprising: At least one memory; and At least one processor, coupled to the at least one memory, is configured to: The input is received via an artificial neural network (ANN), which includes an image and sparse depth measurements; The ANN is used to extract multi-scale visual features from the input; The ANN is used to apply a self-attention mechanism to the multi-scale visual features to generate a set of attention-processed multi-scale visual features; and The ANN generates a dense depth map based on the attention-processed multi-scale visual features of the group.

[0087] 11. The apparatus according to Clause 10, wherein the sparse depth measurement includes light detection and ranging (LiDAR) measurement, red-green-blue depth (RGBD) measurement, or time-of-flight (ToF) measurement.

[0088] 12. The apparatus according to Clause 10 or 11, wherein the at least one processor is included in a mobile device.

[0089] 13. The apparatus according to any one of Clauses 10 to 12, wherein the at least one processor is further configured to implement the dense depth map in an extended reality (XR) application, an autonomous driving application, a robotics application, or an image processing application.

[0090] 14. The apparatus according to any one of clauses 10 to 13, wherein the at least one processor is further configured to process the multi-scale visual features by applying depthwise separable convolutions to a subset of the multi-scale visual features via the ANN.

[0091] 15. The apparatus according to any one of Clauses 10 to 14, wherein the ANN comprises a sparse-to-dense (S2D) network.

[0092] 16. The apparatus according to any one of clauses 10 to 15, wherein the ANN comprises a convolutional neural network (CNN).

[0093] 17. The apparatus according to any one of Clauses 10 to 16, wherein the image is captured by a single camera.

[0094] 18. The device according to any one of clauses 10 to 17, wherein the at least one processor and the single camera are included in a mobile device.

[0095] 19. A non-transitory computer-readable medium having program code recorded thereon, the program code being executed by at least one processor and comprising: Program code for receiving input via an artificial neural network (ANN), the input including images and sparse depth measurements; Program code for extracting multi-scale visual features from the input using the ANN; Program code for applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features; and Program code for generating dense depth maps using the ANN based on the group of attention-processed multi-scale visual features.

[0096] 20. The non-transitory computer-readable medium as described in Clause 19, wherein the sparse depth measurement includes light detection and ranging (LiDAR) measurement, red-green-blue depth (RGBD) measurement, or time-of-flight (ToF) measurement.

[0097] 21. The non-transitory computer-readable medium according to Clause 19 or 20, wherein the program code is included in a mobile device and the program code is executed by the at least one processor of the mobile device.

[0098] 22. The non-transitory computer-readable medium according to any one of Clauses 19 to 21 further includes program code for implementing the dense depth map in extended reality (XR) applications, autonomous driving applications, robotic applications, or image processing applications.

[0099] 23. A non-transitory computer-readable medium according to any one of clauses 19 to 22, wherein the program code includes program code for processing the multi-scale visual features by applying depthwise separable convolutions to a subset of the multi-scale visual features via the ANN.

[0100] 24. The non-transitory computer-readable medium according to any one of Clauses 19 to 23, wherein the ANN comprises a sparse-to-dense (S2D) network.

[0101] 25. A non-transitory computer-readable medium according to any one of Clauses 19 to 24, wherein the image is captured by a single camera.

[0102] 26. An apparatus comprising: A component for receiving input via an artificial neural network (ANN), the input including images and sparse depth measurements; Components used to extract multi-scale visual features of the input through the ANN; A component for applying a self-attention mechanism to the multi-scale visual features via the ANN to generate a set of attention-processed multi-scale visual features; and A component for generating dense depth maps using the ANN based on the group of attention-processed multi-scale visual features.

[0103] 27. The apparatus according to Clause 26, wherein the sparse depth measurement includes light detection and ranging (LiDAR) measurement, red-green-blue depth (RGBD) measurement, or time-of-flight (ToF) measurement.

[0104] 28. The device as described in Clause 26 or 27, wherein the device includes a mobile device.

[0105] 29. The apparatus according to any one of Clauses 26 to 28, the apparatus further comprising components for implementing the dense depth map in extended reality (XR) applications, autonomous driving applications, robotic applications, or image processing applications.

[0106] 30. The apparatus according to any one of clauses 26 to 29, the apparatus further comprising means for processing the multi-scale visual features by applying depthwise separable convolutions to a subset of the multi-scale visual features via the ANN.

[0107] 31. The apparatus according to any one of Clauses 26 to 30, wherein the ANN comprises a sparse-to-dense (S2D) network.

[0108] 32. The apparatus according to any one of Clauses 26 to 31, wherein the image is captured by a single camera.

[0109] In some aspects, the receiving, extracting, applying, and / or generating components may be CPU 102, program memory associated with CPU 102 / 422, dedicated memory block 118, fully connected layer 362, GPU 104 / 426, NPU 108 / 428, and / or routing connection processing unit 216, configured to perform the described functions. In another configuration, the aforementioned components may be any module or device configured to perform the functions described by the aforementioned components.

[0110] The various operations of the methods described above can be performed by any suitable component capable of performing the corresponding function. These components may include various hardware and / or software components and / or modules, including but not limited to circuits, application-specific integrated circuits (ASICs), or processors. Generally, in the cases where operations are illustrated in the accompanying drawings, these operations may have corresponding paired components with similar numbering plus functional components.

[0111] As used, the term "determine" encompasses a wide variety of actions. For example, "determine" can include calculation, operation, processing, deduction, investigation, lookup (e.g., searching in a table, database, or other data structure), assertion, etc. Additionally, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Furthermore, "determine" can include parsing, selecting, choosing, building, etc.

[0112] As used, the phrase "at least one of the items in the list" refers to any combination of these items, including a single member. As an example, "at least one of a, b, or c" is intended to cover: a, b, c, ab, ac, bc, and abc.

[0113] The various exemplary logic blocks, modules, and circuits described in this disclosure can be implemented or executed using a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic component, discrete hardware component, or any combination thereof designed to perform the described functions. While the general-purpose processor may be a microprocessor, in alternative embodiments, the processor may be any commercially available processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration.

[0114] The steps or algorithms of the methods described in this disclosure may be directly embodied in hardware, a software module executed by a processor, or a combination of both. The software module may reside in any form of storage medium known in the art. Some examples of usable storage media include random access memory (RAM), read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs, and the like. The software module may include a single instruction or multiple instructions and may be distributed across several different code segments, across different programs, and across multiple storage media. The storage medium may be coupled to the processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium may be integral with the processor.

[0115] The disclosed method includes one or more steps or actions for implementing the described method. The steps and / or actions of the method may be interchanged without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of a particular step and / or action may be modified without departing from the scope of the claims.

[0116] The described functionality can be implemented in hardware, software, firmware, or any combination thereof. If implemented in hardware, an example hardware configuration may include a processing system within the device. This processing system may utilize a bus architecture. Depending on the specific application and overall design constraints of the processing system, the bus may include any number of interconnect buses and bridges. The bus can link various circuits together, including processors, machine-readable media, and bus interfaces. The bus interface can be used to connect network adapters, etc., to the processing system via the bus. The network adapter can be used to implement signal processing functions. In some respects, user interfaces (e.g., keypads, displays, mice, joysticks, etc.) may also be connected to the bus. The bus may also link various other circuits, such as timing sources, peripherals, voltage regulators, power management circuits, etc., which are well known in the art and will not be described further.

[0117] A processor may be responsible for managing the bus and general-purpose processing, including executing software stored on a machine-readable medium. A processor may be implemented using one or more general-purpose processors and / or special-purpose processors. Examples include microprocessors, microcontrollers, DSP processors, and other circuitry that can execute software. Software should be interpreted broadly as instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. By way of example, a machine-readable medium may include random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, disks, optical disks, hard disks, or any other suitable storage medium, or any combination thereof. A machine-readable medium may be embodied as a computer program product. A computer program product may include packaging material.

[0118] In a hardware implementation, machine-readable media can be part of a processing system separate from the processor. However, as those skilled in the art will readily understand, machine-readable media, or any portion thereof, can be external to the processing system. By way of example, machine-readable media may include transmit lines, carrier waves modulated by data, and / or computer components separate from the device, all accessible to the processor via a bus interface. Alternatively or additionally, machine-readable media, or any portion thereof, may be integrated into the processor, such as in the case of a cache and / or a general-purpose register file. Although the various components discussed may be described as having a specific location, such as local components, they may also be configured in various ways, such as certain components being configured as part of a distributed computing system.

[0119] The processing system may be configured as a general-purpose processing system having one or more microprocessors providing processor functionality and external memory providing at least a portion of machine-readable medium, all of which are linked together with other supporting circuitry via an external bus architecture. Alternatively, the processing system may include one or more neuromorphic processors for implementing the described neuron and nervous system models. As another alternative, the processing system may be implemented using an application-specific integrated circuit (ASIC) having a processor, bus interface, user interface, supporting circuitry, and at least a portion of machine-readable medium integrated on a single chip, or using one or more field-programmable gate arrays (FPGAs), programmable logic devices (PLDs), controllers, state machines, gated logic components, discrete hardware components, or any other suitable circuitry, or any combination of circuitry capable of performing the various functionalities described throughout this disclosure. Those skilled in the art will recognize how best to implement the described functionality of the processing system depends on the specific application and the overall design constraints imposed on the system as a whole.

[0120] Machine-readable media may include multiple software modules. These software modules include instructions that, when executed by a processor, enable the processing system to perform various functions. Software modules may include send and receive modules. Each software module may reside in a single storage device or be distributed across multiple storage devices. For example, when a triggering event occurs, a software module may be loaded from a hard disk drive into RAM. During the execution of a software module, the processor may load some of the instructions into a cache to improve access speed. One or more cache lines may then be loaded into a general-purpose register file for processor execution. When the functionality of a software module is referred to below, it will be understood that such functionality is implemented by the processor when executing the instructions from that software module. Furthermore, it should be understood that aspects of this disclosure result in improvements to the functionality of a processor, computer, machine, or other system implementing such aspects.

[0121] If implemented in software, the functions may be stored as one or more instructions or codes on or transmitted through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one location to another. A storage medium can be any available medium accessible to a computer. By way of example and not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and is accessible to a computer. Additionally, any connection is also appropriately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, optical fiber, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared (IR), radio, and microwave, then such coaxial cable, optical fiber, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. The disks and optical discs used include compact discs (CDs), laser discs, optical discs, digital multifunction discs (DVDs), floppy disks, and Blu-ray discs. ® Optical discs, where magnetic disks typically reproduce data magnetically, and optical discs reproduce data optically using lasers. Therefore, in some aspects, computer-readable media may include non-transitory computer-readable media (e.g., tangible media). Furthermore, in other aspects, computer-readable media may include transient computer-readable media (e.g., signals). Combinations of the above should also be included within the scope of computer-readable media.

[0122] Therefore, certain aspects may include a computer program product for performing the presented operations. For example, such a computer program product may include a computer-readable medium on which instructions are stored (and / or encoded) that can be executed by one or more processors to perform the described operations. In some aspects, the computer program product may include packaging material.

[0123] Furthermore, it should be understood that modules and / or other suitable components for performing the described methods and techniques may be downloaded and / or otherwise obtained by the user terminal and / or base station where applicable. For example, such devices can be coupled to a server to facilitate the delivery of components for performing the described methods. Alternatively, the various methods described can be provided via storage components (e.g., RAM, ROM, physical storage media such as CDs or floppy disks) so that the user terminal and / or base station can obtain the various methods once the storage component is coupled to or provided to the device. In addition, any other suitable techniques suitable for providing the described methods and techniques to the device may be utilized.

[0124] It should be understood that the claims are not limited to the precise configurations and components illustrated above. Various modifications, variations, and alterations may be made to the arrangement, operation, and details of the methods and apparatus described above without departing from the scope of the claims.

Claims

1. A processor-implemented method performed by at least one processor, the processor-implemented method comprising: receiving, by an artificial neural network (ANN), an input comprising an image and sparse depth measurements; extracting, by the ANN, multiscale visual features of the input; applying, by the ANN, a self-attention mechanism to the multiscale visual features to generate a set of attention-processed multiscale visual features; and generating, by the ANN, a dense depth map based on the set of attention-processed multiscale visual features.

2. The processor-implemented method of claim 1, wherein the sparse depth measurements comprise light detection and ranging (LiDAR) measurements, red-green-blue depth (RGBD) measurements, or time-of-flight (ToF) measurements.

3. The processor-implemented method of claim 1, wherein the processor-implemented method is performed by at least one processor of a mobile device.

4. The processor-implemented method of claim 1, further comprising implementing the dense depth map in an extended reality (XR) application, an autonomous driving application, a robotics application, or an image processing application.

5. The processor-implemented method of claim 1, further comprising processing, by the ANN, the multiscale visual features by applying depthwise separable convolutions to a subset of the multiscale visual features.

6. The processor-implemented method of claim 1, wherein the ANN comprises a sparse-to-dense (S2D) network.

7. The processor-implemented method of claim 1, wherein the ANN comprises a convolutional neural network (CNN).

8. The processor-implemented method of claim 1, wherein the image is captured by a single camera.

9. The processor-implemented method of claim 8, wherein the processor-implemented method is performed by at least one processor of a mobile device, wherein the single camera is included in the mobile device.

10. An apparatus, the apparatus comprising: at least one memory; and at least one processor coupled to the at least one memory, the at least one processor configured to: receive, by an artificial neural network (ANN), an input comprising an image and sparse depth measurements; extract, by the ANN, multiscale visual features of the input; apply, by the ANN, a self-attention mechanism to the multiscale visual features to generate a set of attention-processed multiscale visual features; and generate, by the ANN, a dense depth map based on the set of attention-processed multiscale visual features.

11. The apparatus of claim 10, wherein the sparse depth measurements comprise light detection and ranging (LiDAR) measurements, red-green-blue depth (RGBD) measurements, or time-of-flight (ToF) measurements.

12. The apparatus of claim 10, wherein the at least one processor is included in a mobile device. ​ ​ 13. The apparatus of claim 10, wherein the at least one processor is further configured to implement the dense depth map in an extended reality (XR) application, an autonomous driving application, a robotics application, or an image processing application.

14. The apparatus of claim 10, wherein the at least one processor is further configured to process the multi-scale visual features by the ANN by applying depth-wise separable convolutions to a subset of the multi-scale visual features.

15. The apparatus of claim 10, wherein the ANN comprises a sparse-to-dense (S2D) network.

16. The apparatus of claim 10, wherein the ANN comprises a convolutional neural network (CNN).

17. The apparatus of claim 10, wherein the image is captured by a single camera.

18. The apparatus of claim 17, wherein the at least one processor and the single camera are included in a mobile device.

19. A non-transitory computer-readable medium having program code recorded thereon, the program code executed by at least one processor and comprising: program code to receive, by an artificial neural network (ANN), an input comprising an image and sparse depth measurements; program code to extract, by the ANN, multi-scale visual features of the input; program code to apply, by the ANN, a self-attention mechanism to the multi-scale visual features to generate a set of attention-processed multi-scale visual features; and program code to generate, by the ANN, a dense depth map based on the set of attention-processed multi-scale visual features.

20. The non-transitory computer-readable medium of claim 19, wherein the sparse depth measurements comprise light detection and ranging (LiDAR) measurements, red-green-blue depth (RGBD) measurements, or time-of-flight (ToF) measurements.

21. The non-transitory computer-readable medium of claim 19, wherein the program code comprises program code to process, by the ANN, the multi-scale visual features by applying depth-wise separable convolutions to a subset of the multi-scale visual features.

22. The non-transitory computer-readable medium of claim 19, wherein the ANN comprises a sparse-to-dense (S2D) network.

23. The non-transitory computer-readable medium of claim 19, wherein the image is captured by a single camera.

24. An apparatus comprising: means for receiving, by an artificial neural network (ANN), an input comprising an image and sparse depth measurements; means for extracting, by the ANN, multi-scale visual features of the input; means for applying, by the ANN, a self-attention mechanism to the multi-scale visual features to generate a set of attention-processed multi-scale visual features; and means for generating, by the ANN, a dense depth map based on the set of attention-processed multi-scale visual features. ​ ​ 25. The apparatus of claim 24, wherein the sparse depth measurements comprise light detection and ranging (LiDAR) measurements, red-green-blue depth (RGBD) measurements, or time-of-flight (ToF) measurements.

26. The apparatus of claim 24, further comprising means for processing the multiscale visual features by the ANN by applying depthwise separable convolutions on a subset of the multiscale visual features.

27. The apparatus of claim 24, wherein the ANN comprises a sparse-to-dense (S2D) network.

28. The apparatus of claim 24, wherein the image is captured by a single camera.

Images