44 results about "Input format" patented technology

The invention discloses a method for automatically drawing an OpenGL (Open Graphics Library) program by using Vulkan. The method comprises the following steps of: creating a context used by the Vulkan, initializing each module, processing an OpenGL instruction related to texture and data buffering, and managing storage of texture and data buffering resources in a video memory; a shader program used by the OpenGL is preprocessed into a format acceptable to Vulkan, and an OpenGL shader program instruction is created and destroyed; processing an OpenGL (Open Graphics Library) instruction related to frame buffering to generate structural body information required by Vulkan dynamic rendering; an OpenGL instruction of the sampler is also created; processing an OpenGL (Open Graphics Library) instruction for creating a vertex input format and managing a vertex data buffer area, and maintaining vertex input information, a vertex buffer area and an index buffer area required by Vulkan; and finally, drawing or calculating, distributing and calling Vulkan on the basis of all the instructions.

The invention discloses a short video production method and system based on artificial intelligence, and relates to the technical field of multimedia processing, and the method comprises the steps: inputting a split script into a multi-modal generation engine, employing a lower triangle space-time attention matrix to generate a visual stream, and generating a text stream and an audio stream; calculating a neural conduction delay amount between the visual stream and the audio stream based on an audio-visual perception delay prediction model, and generating a corrected audio stream through time axis forward compensation; calculating a cross-modal attention weight matrix according to the node association strength of the dynamic causal graph, and projecting the text stream and the corrected audio stream to a joint feature space to generate a multi-modal fusion feature; and inputting the multi-modal fusion features into a format adaptation engine, and after time-space consistency verification, embedding a neural implicit watermark and encoding the neural implicit watermark into a short video. According to the invention, the semantic consistency and coordination among the text, the audio and the visual content are improved, so that high-quality and high-intelligence automatic generation of the short video is realized on the whole.

The invention provides an automatic cross-project vulnerability verification method and device based on a large language model, relates to the technical field of vulnerability verification, and aims to solve the problem that whether an upstream vulnerability can still be triggered or not is difficult to efficiently confirm in the prior art in a scene that significant differences exist between parameter semantics and input formats of a source project and a target project. According to the method, a multi-agent parallel target sensing parameter migration framework is constructed, and a legal command line parameter combination related to a vulnerability function is automatically generated from a target software manual in combination with an RAG retrieval enhancement technology; a ReAct intelligent agent is further introduced, a source PoC file is embedded into an acceptable input shell of target software in a cross-format mode under the condition that manual intervention is not needed, and an initial cross-format seed is formed. And iteratively optimizing the parameter-seed pair by taking the function-level trajectory similarity as a feedback index, and driving the parameter sensitive grey box fuzzy test to quickly converge to accurate input capable of triggering vulnerabilities.

The invention relates to the field of high-performance computing cluster resource scheduling management, and discloses a Slurm scheduling specification integration method and system, and the method comprises the following steps: 1, analyzing a heterogeneous job description file, and extracting a resource demand parameter and a dependency relationship through a regular expression rule base; 2, based on the extracted original parameters, converting the original parameters into SLURM standard parameters through a preset mapping rule; step 3, according to the converted standard parameters; 4, calling a Slurm interface command to submit a script; and step 5, monitoring the execution state of the submitted job, and triggering a re-submission process for the abnormal job with the resource overrun or dependency missing. Through a multi-level analysis architecture and a regular expression rule base, job description files in different formats are effectively compatible, the problem that analysis of a non-standardized input format by a traditional method fails is solved, unified processing of cross-platform job definition is achieved, and the manual adaptation cost is remarkably reduced.

The invention discloses an active optical alignment method based on a local lightweight language model, and particularly relates to the technical field of precision installation and adjustment, comprising the following steps: extracting image indexes and platform attitude information through image acquisition, constructing a unified input format, verifying and inputting the language model; generating a six-degree-of-freedom pose adjustment strategy and a credibility score; the strategy is executed by the platform control device after being verified by a preset criterion; scoring is performed again after execution, and whether a feature posture is recorded or error compensation is triggered is judged according to a scoring result, so that subsequent input and strategy generation are optimized, and alignment precision and stability are improved; according to the method, the data consistency is guaranteed through a unified input format and a verification mechanism, the alignment precision is improved in combination with a dynamic scoring strategy, and the stability and fault tolerance of the system are enhanced by introducing a feature attitude recording and error back-filling mechanism, so that a strategy generation and execution process with high reliability, high adaptation and high robustness in an active optical alignment task is realized.

The invention relates to an access control strategy automatic generation system and method based on multi-task learning, and the method comprises the steps: carrying out the word segmentation, cleaning and embedded vector conversion of an original access control text through a data preprocessing module, and constructing a normative input format; the feature sharing layer module is used for extracting deep semantic features of a text through multi-layer bidirectional coding and an attention mechanism and providing unified representation for downstream tasks; the access control statement identification module is used for judging whether each sentence in the text is an access control statement or not and realizing automatic identification of strategy related contents; and the attribute extraction and annotation module is used for annotating words in the access control statements and extracting subject, object and operation access control attributes. A word coding layer and a sentence coding layer are shared, local and global attention mechanisms are combined, key information of a text is extracted, the semantic understanding ability is enhanced, and cooperative training of statement recognition and attribute extraction is achieved; a conditional random field CRF structure is used for sequence labeling, and the structural rationality of attribute labels is ensured.

The invention provides an intelligent form filling method and device for a browser plug-in based on multi-modal interaction, and relates to the technical field of data processing.The method comprises the steps that to-be-filled information and a to-be-filled form are obtained, and the to-be-filled information has multiple input formats; converting the to-be-filled information into a filling field in a text format; converting the form to be filled into a structured form, wherein the structured form has a filling position corresponding to the filling field; filling the filling field into the corresponding filling position to obtain a filling template; and matching a corresponding backfilling position for the content in the filling template in the to-be-filled form, and filling the content in the filling template into the corresponding backfilling position to obtain a complete form. According to the intelligent filling method, after the to-be-filled information and the to-be-filled form input by the user are received, the to-be-filled information can be automatically filled into the to-be-filled form, the manual filling process is reduced, and the filling efficiency is improved.

Provided are a feature encoding / decoding method and device, a recording medium on which a bitstream generated by the feature encoding method is stored, and a method for transmitting the bitstream. The feature decoding method according to the present disclosure comprises the steps of: determining whether or not an input source is feature data; determining a feature map type of the feature data based on the input source being the feature data; obtaining encoding information of the feature data from a bitstream; and decoding the feature data based on the feature map type and the encoding information, and whether or not the input source is the feature data may be determined based on input format identification information that is obtained from a sequence parameter set.

Disclosed herein are methods and systems for generating metadata from content using one or more machine learning models. In an embodiment, a method may include receiving the content through a graphical user interface associated with the large language model, generating a first file by tokenizing the content into an input format for the large language model and merging the tokenized content with a content instruction, inputting the first file into the large language model, generating, using the large language model, metadata from at least the first file, the metadata reflecting a context associated with the content, generating a second file, the second file comprising the metadata, and displaying the generated metadata on the graphical user interface.

The invention relates to the technical field of advertisement material auditing methods, in particular to an advertisement material auditing method and device and a storage medium, and the method specifically comprises the following steps: 1, carrying out the standardized collection of multi-source advertisement materials: firstly, carrying out the unified collection of the advertisement materials from different delivery platforms, the types of the collected materials comprise image materials, video materials, text materials and audio materials, and the materials in different formats are converted into a unified auditing input format by adopting a standardized processing algorithm; according to the method, through the seven steps of multi-source material standardized collection, multi-dimensional feature extraction, feature weight dynamic allocation, preliminary screening compliance judgment, secondary screening multi-modal fusion auditing, manual rechecking triggering and model iteration updating, compared with manual auditing, through machine automatic processing, the auditing efficiency is improved, and the corresponding comprehensive analysis ability is achieved.

A mechanism for processing video data is disclosed. The mechanism includes determining that a value of a neural-network post-filter characteristics (NNPFC) input format indicator (nnpfc_inp_format_inc) is coded as a u(N) coded syntax element where N is an integer greater than 0. A conversion is performed between a visual media data and a bitstream based on the NNPFC input format indicator.

The invention provides a friction stir welding defect detection method fused with weak supervised learning, which belongs to the technical field of welding defect detection, and constructs a unified detection framework suitable for various weak labels such as points, frames, graffiti and the like by referring to the advantages of a large model SAM (Section Anything Model) in visual segmentation. The method comprises the following steps: firstly, converting different types of weak tags into an input format acceptable by SAM by adopting a prompt adapter module, and enhancing the prompt compatibility of a model; and secondly, abnormal region responses are eliminated through a response filter, and the recognition precision of the disguise or low-contrast defect region is improved in combination with a semantic matcher. A prompt self-adaptive knowledge distillation mechanism is further introduced, so that knowledge migration from the SAM model to the lightweight detection model is realized, and the feature learning ability of complex weld defects is enhanced. According to the method, high-quality defect detection can be realized in the weld defect image without a large number of accurate labels, and particularly, the method has remarkable advantages in the aspect of processing the welding defects with fuzzy boundaries, small sizes or weak contrast, and has wide industrial application value.

The invention provides a VLA model safety testing method based on characteristic disturbance, and belongs to the technical field of VLA model safety testing. In order to solve the technical problems of strong perceptibility and insufficient characteristic disturbance of the existing model robustness test method in a multi-modal model, the adopted technical scheme is as follows: carrying out standardization processing on an input image to enable the input image to accord with an input format of a visual encoder; obtaining internal multi-layer value representation of the model on an undisturbed image as an attack reference target; generating confrontation disturbance by using a generative network, and enabling the final feature expression to generate obvious offset; obtaining the internal multilayer value representation of the model on the disturbance image, obtaining the internal representation of the disturbance image in the visual encoder, and evaluating the attack effect; constructing a confrontation loss function to maximize the difference between the original feature and the disturbance feature, introducing a disturbance regular term to limit the disturbance intensity, and training to obtain a test model; the method is applied to VLA model security testing.

A mechanism for processing video data is disclosed. The mechanism includes determining that a value of a neural network post-processing filter characteristic (NNPFC) input format indicator (nnpfcinformatinc) is within a range from 0 to N (including a boundary value), where N is a positive integer. Conversion between the visual media data and the bitstream is performed based on the NNPFC input format indicator.

A mechanism for processing video data is disclosed. The mechanism includes determining a value of a neural network post-processing filter characteristic (NNPFC) input format indicator (nnpfcinformatinc) to be encoded as a syntax element for u (N) encoding, where N is an integer greater than 0. Conversion between the visual media data and the bitstream is performed based on the NNPFC input format indicator.

The invention relates to a picture retrieval method and device based on understanding. The method comprises the following steps: processing a to-be-retrieved picture by utilizing a picture encoder, and outputting a picture vector required for understanding a large language model; finely adjusting the reply format and the input format of the large language model according to the requirement of the retrieval scene; the big language model understands an input picture vector, and outputs an analysis result containing related elements in a picture and an understanding semantic text of the picture; performing target identification on related elements in the picture according to the understanding semantic text of the picture, performing type identification on an identified target, and then performing target matching; and for the matched target, extracting a corresponding picture vector, and judging whether the picture is abnormal or not according to the understanding text description information of the picture. According to the scheme, the pictures are associated with the text semantics, understanding weighting is carried out on the properties, key attributes and types of the pictures through the semantics, the flexibility is high, the universality is high, and the retrievable content is richer and more accurate.

A content preparation method, apparatus, device, and storage medium, the method comprising receiving content prepared for media streaming or a media streaming network, the content corresponding to an input Common Media Application Format (CMAF) track and a set of output CMAF tracks; determining a Content Preparation Template (CPT), wherein the CPT includes a Network-based Media Processing (NBMP) workflow description document (WDD), the NBMP WDD specifying the input format of the input CMAF track and an array of task instances corresponding to the set of output CMAF tracks; preparing the content according to the CPT; and streaming the prepared content to a media streaming client via a media streaming network.

The invention discloses an input length determination method and device of a question and answer model, equipment and a medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: according to a preset long text of at least one language type, determining a candidate long text of at least one input format corresponding to the preset long text of each language type; performing text combination on the preset answer text and the candidate long text to obtain at least one candidate input text corresponding to the candidate long text of each input format; based on the target question and answer model, according to a preset question and answer instruction and the candidate input text, iteratively optimizing the text length of the candidate long text in the candidate input text to obtain at least one candidate input length of the target question and answer model; and determining a reliable input length of the target question and answer model according to the at least one candidate input length. According to the technical scheme, the reliable input length of the target question and answer model is explored through the combination mode of multiple answers and texts, and the reliability of determining the input length is improved.

The application discloses a patent text abstract extraction method, system and device and a storage medium. In the application, the patent text of a patent to be extracted is acquired; text embedding is performed on the patent text to obtain embedded text; the text data is converted into embedded text conforming to the input format of a subsequent model, so that the analysis efficiency of the model is improved; an abstract extraction model is used to perform abstract analysis on the embedded text to obtain abstract text; the abstract extraction model is obtained through combined cycle training of an expansion model, and the abstract model and the expansion model are both generated based on a BERT model; the expansion model is used to perform expansion analysis on the abstract text to obtain expansion text; through combined training of the abstract extraction model and the expansion model, the purpose of model training is strengthened, the accuracy of patent abstract extraction is effectively improved, and the application can be widely used in the technical field of text semantic processing.

The invention relates to the technical field of lip language recognition, and discloses a lightweight lip language recognition model based on multi-teacher adaptive distillation, and the model comprises the steps: 1, carrying out the face detection, lip region-of-interest cutting and alignment of an input lip language video sequence, and obtaining a lip region-of-interest sequence; performing composite enhancement of directional design on the lip region-of-interest sequence; 2, extracting spatial-temporal features from the lip region-of-interest sequence by using a three-dimensional residual network variant to obtain unified sequence features adaptive to input formats of a multi-teacher model and a lightweight student model; and step 3, constructing a heterogeneous multi-teacher network. The technical scheme of heterogeneous multi-teacher network simultaneous prediction and weighted knowledge fusion is adopted, the technical effect of fully utilizing the complementary advantages of different teacher structures is achieved, knowledge breadth and depth synchronous migration is achieved, and the defects that the knowledge coverage range of a single teacher is limited and the generalization ability of a student model is insufficient are overcome.

The invention belongs to the technical field of transaction, and particularly relates to an intelligent market transaction decision-making method and system based on a large language model. According to the system, by combining multi-modal data (including time sequence data and text data) and adopting a multi-stage decision processing framework, the limitation of a traditional transaction system in processing mass data, information overload and market environment dynamic change is effectively solved. Specifically, the system comprises a plurality of modules: a multi-modal information sensing module used for collecting and processing market data and news texts; the time series data encoding module converts time series data such as market price and transaction volume into a large language model input format suitable for natural language processing; the historical data storage and retrieval module is combined with short-term and long-term market information to improve the prediction capability of the system; and the adaptive decision-making module based on the technical indexes is used for coping with different market conditions through dynamic adjustment strategies.

The invention relates to a sparse matrix feature extraction method and system based on multi-format input and parallel computing, and the method comprises the steps: obtaining a to-be-processed sparse matrix from a sparse matrix file supporting multiple input formats, carrying out the format analysis, and obtaining global meta-information and a unified data view; according to the global meta-information and the unified data view, loading and dividing the sparse matrix in an MPI distributed environment to obtain distributed division information and a plurality of process local sub-blocks; according to the global meta-information and the distributed division information, non-zero elements of local sub-blocks of the processes are traversed in parallel by utilizing OpenMP multithreading in the local sub-blocks of the processes, local statistical features are determined, and a local imagination feature matrix with a fixed resolution is constructed; according to the global meta-information, reduction and aggregation are carried out on the local statistical features and the local imagination feature matrixes of the local sub-blocks of all processes through MPI communication, global unified global statistical features and global imagination feature matrixes are obtained, and the processing efficiency and format compatibility are improved.

A device for encoding audio object parameters, the device comprising means for: identifying one of at least two audio objects to be individually encoded within an audio environment; obtaining a ratio parameter of the at least two audio objects for a time-frequency element of a frame comprising at least one time element and at least one frequency element; grouping, for the time-frequency element, a ratio parameter associated with the at least two audio objects, quantizing the ratio parameter within the packet, where the quantizing the ratio parameter is configured to generate an integer representation of ratio parameter values that are summed to a defined integer value for a particular packet; encoding the integer representation of the ratio parameter values within a first packet as an enumerated index; a subsequent packet is encoded as a difference relative to the first packet.

The invention discloses a small-sample text classification method based on large model cue word driving, and the method comprises the steps: constructing a structured cue word template which comprises task description, classification label detailed rules, input format description, decision steps and small-sample examples; k samples most similar to target text semantics are retrieved through a vector database to serve as dynamic examples, and retrieval is achieved based on cosine similarity; inputting the structured cue word template and the dynamic example combination into a large language model; and analyzing the structured label output by the large language model to finish text classification. And a large model is guided to efficiently complete a classification task through well-designed cue word templates and examples. Furthermore, the classification result generated by the large model is used for fine adjustment of the lightweight small model, the small model can better inherit the classification capability of the large model, the reasoning cost is greatly reduced while high precision is kept, and a solution with both efficiency and practicability is provided for automatic classification in a data scarcity scene.

A mechanism for processing video data is disclosed. The mechanism includes determining that a value of a neural-network post-filter characteristics (NNPPC) input format indicator (nnpfc_inp_format_inc) is in a range of 0 to N, inclusive, where N is a positive integer. A conversion is performed between a visual media data and a bitstream based on the NNPPC input format indicator.

This invention discloses a multimodal benchmark evaluation method for large language models. First, structured information is extracted from the original PPT file, and a bimodal input format including structured JSON data and slide screenshots is constructed to ensure that the large model can simultaneously acquire structured layout information and complete visual information. A multi-level task module is constructed, including detection, understanding, modification, and generation tasks. For each of the constructed detection, understanding, modification, and generation tasks, corresponding evaluation indicators and calculation methods are designed to achieve reproducible evaluation of the large model's capabilities. This evaluation method can systematically characterize the differences in capabilities of large models in slide element recognition, layout reasoning, structured editing, and visual design generation, and can serve as an important benchmark for multimodal large models in the field of office document processing.

This invention provides a method and system for lossless upscaling of low-resolution images based on a large AI model. The method includes: parallel image feature extraction from received images of different formats using a large AI model; determining a pixel generation strategy based on the feature extraction results, a pixel condition generation mechanism, and image content; performing resolution processing on the images based on the pixel generation strategy; filtering noise from the images based on the resolution processing results; automatically recommending an optimal upscaling ratio based on the noise filtering results and the image resolution; upscaling the images based on the optimal upscaling ratio; performing personalized adjustments based on user needs after upscaling; and exporting the personalized adjustments based on the input format. This method ensures that image quality is significantly maintained while upscaling low-resolution images, greatly improving image processing efficiency and guaranteeing lossless upscaling of low-resolution images.

In a video coding and decoding method, in a case that a transmission rate is low, image quality is deteriorated due to coding distortion, but the image quality can be improved owing to post-filter processing using a neural network. However, a problem is that there may be no method of defining a strength control value to be input in the post-filter processing. A video decoding apparatus according to an aspect of the present invention includes an image decoding apparatus that generates a decoded image, and a post-filter processing apparatus that performs post-filter processing on the decoded image. In accordance with an input format for luminance and chrominance of the decoded image to be input in the post-filter processing, a strength control value obtained in a process of generating the decoded image is transformed, and the post-filter processing is performed.

Negotiation of a conversational immersive audio session includes obtaining one or more supported immersive conversational codec input formats, providing the one or more immersive conversational codec input formats in an order as a list, including an immersive conversational codec input format attribute in a session description file, populating the input format attribute with the list of immersive conversational codec input formats in the session description file, generating a session negotiation offer, based on the session description file, and transmitting the session negotiation offer to a receiver user equipment.

Negotiation of a dialog immersive audio session includes obtaining one or more supported immersive dialog codec input formats; providing one or more immersive dialogue codec input formats as a list according to a certain sequence; the input format attribute of the immersive dialogue codec is included in the session description file; populating the input format attributes with a list of immersive dialogue codec input formats in the session description file; generating a session negotiation proposal based on the session description file; and sending the session negotiation proposal to the recipient user equipment. Non-limiting example embodiments may include receiving a session negotiation proposal; parsing one or more immersive dialog codec input formats from a session description file in the received session negotiation proposal; obtaining one or more supported and preferred input formats of the recipient user equipment; selecting, from the received session negotiation proposal, one or more preferred input formats of the sender user equipment that are common to the one or more supported and preferred input formats of the receiver user equipment; populating the immersive dialog codec input format attributes with the selected one or more preferred input formats within the session negotiation response; and sending a session negotiation response to the sender user equipment.