Alerta

UNSUPERVISED LEARNING OF OBJECT REPRESENTATIONS FROM VIDEO SEQUENCES USING ATTENTION OVER SPACE AND TIME

Resumen de: US20260187982A1

A computer-implemented video generation neural network system, configured to determine a value for each of a set of object latent variables by sampling from a respective prior object latent distribution for the object latent variable. The system comprises a trained image frame decoder neural network configured to, for each pixel of each generated image frame and for each generated image frame time step process determined values of the object latent variables to determine parameters of a pixel distribution for each of the object latent variables, combine the pixel distributions for each of the object latent variables to determine a combined pixel distribution, and sample from the combined pixel distribution to determine a value for the pixel and for the time step.

DECISION TREE OF MODELS: USING DECISION TREE MODEL, AND REPLACING THE LEAVES OF THE TREE WITH OTHER MACHINE LEARNING MODELS

Resumen de: US20260187484A1

Described are techniques of generating and training a neural network that include training multiple models and constructing multiple decision trees with said models. Each decision tree may include additional decision trees at various levels of that decision tree. Each decision tree has a different accuracy indicator due to the unique structuring of each decision tree, and by testing each tree through a testing dataset, the tree with the highest accuracy can be determined.

METHOD FOR TRACKING A DENTAL MOVEMENT

Resumen de: US20260183087A1

Determine an amount of movement between an “anterior” updated instant and a “posterior” updated instant after the anterior updated instant. At the anterior updated instant, implementing an analysis method to obtain an anterior updated item of spatial information. Analyze a set of several images to obtain spatial information. Create a historical learning database relating to a dental body and the spatial attribute, with a set of historical images all depicting the dental body and an item of spatial information.Training the neural network, by providing the sets of several historical images as input and with the historical spatial information as output, with the spatial attribute defining an ordered sequence of variables in a three-dimensional reference frame. At the posterior updated instant, implement analysis to obtain a posterior updated item of spatial information. Compare the anterior and posterior updated spatial information, to obtain an amount of movement.

APPARATUS FOR OBTAINING MAGNETIC RESONANCE IMAGES BASE ON DEEP LEARNING MODEL AND METHOD OF CONTROLLING THE SAME

Resumen de: US20260186092A1

0000 The present disclosure provides an apparatus for restoring the quality of magnetic resonance images based on a deep learning model and a method of controlling the same. The method includes: obtaining a training image corresponding to each magnetic resonance image by applying at least one of a plurality of elements set in connection with the quality of the magnetic resonance image to a magnetic resonance signal corresponding to the magnetic resonance image; obtaining a training dataset including the magnetic resonance image as label data and the obtained training image as input data matching the label data; and training a neural network model based on the training dataset and context data corresponding to the training image. Obtaining the training image includes distorting the magnetic resonance signal by applying the at least one of the plurality of elements and obtaining the training image based on the distorted magnetic resonance signal.

MEDICAL IMAGE RECONSTRUCTION METHOD BASED ON DEEP LEARNING

Resumen de: WO2026139108A1

The present invention belongs to the interdisciplinary technical field of medical image processing and artificial intelligence. Disclosed is a medical image reconstruction method based on deep learning. The method comprises: acquiring a medical image data set, and labeling corresponding key anatomical structure information; on the basis of a feature encoder and a graph structure processor, constructing a medical image reconstruction model; the feature encoder performing feature extraction on the medical image data set by means of a convolutional neural network, so as to obtain a multi-scale feature map; the graph structure processor constructing an initial landmark point set of the multi-scale feature map by means of a graph attention network, and analyzing topological relationships between landmark points, so as to obtain an optimized landmark point arrangement; on the basis of the optimized landmark point arrangement, constructing an anatomical structure graph; on the basis of the anatomical structure graph, designing an adaptive loss function to train the medical image reconstruction model; and on the basis of the trained medical image reconstruction model, obtaining a corresponding reconstructed image. The present invention achieves the objective of improving the detail representation of a reconstructed image while maintaining the accuracy of an anatomical structure.

UAV PERCEPTION VALIDATION BASED UPON A SEMANTIC AGL ESTIMATE

Resumen de: US20260188129A1

0000 A method for perception validation of an aerial vehicle includes: acquiring an image of a ground area with an onboard camera system of the aerial vehicle, generating an above ground altitude (AGL) estimate with a neural network trained to output the AGL estimate in response to the image fed as an input to the neural network, generating a motion estimate or a position estimate based upon sensor data output from a sensor disposed onboard the aerial vehicle, and cross-validating the motion or position estimate against the AGL estimate.

OBJECT TRACKING PROCESSING METHOD AND APPARATUS, DEVICE, AND MEDIUM

Resumen de: WO2026138395A1

The present invention provides an object tracking processing method and apparatus, a device, and a medium. The processing method comprises: respectively preprocessing a first-frame image and a current image to generate a corresponding first-frame initial image and a corresponding current initial image; performing feature extraction processing on the first-frame initial image and the current initial image by means of a weight-adaptive fusion encoder to generate input feature data; performing deep-level fusion processing on the first-frame initial image and the current initial image; performing feature extraction processing by means of the weight-adaptive fusion encoder to generate deep feature data; performing attention-weighted fusion processing on the input feature data and the deep feature data by means of the weight-adaptive fusion encoder to generate target feature tensor data; and processing the target feature tensor data by means of a fully convolutional neural network to generate a target image. The object tracking processing method and apparatus, the device, and the medium provided by the present invention improve object tracking performance.

TIME-SERIES FORECASTING SYSTEM, DEVICE, AND METHOD

Resumen de: US20260187497A1

A system generates prediction data by processing an input sequence segmented along a time axis based on a predetermined period using neural networks. The neural networks comprise a first neural network configured to apply dilated attention to the input sequence segmented along the time axis based on the predetermined period, and a second neural network configured to apply random partition attention to data arranged along a feature axis.

SYSTEM AND METHOD FOR GENERATING PLAYER TRACKING DATA FROM BROADCAST VIDEO

Resumen de: US20260187816A1

A system and method of generating a player tracking prediction are described herein. A computing system retrieves a broadcast video feed for a sporting event. The computing system segments the broadcast video feed into a unified view. The computing system generates a plurality of data sets based on the plurality of trackable frames. The computing system calibrates a camera associated with each trackable frame based on the body pose information. The computing system generates a plurality of sets of short tracklets based on the plurality of trackable frames and the body pose information. The computing system connects each set of short tracklets by generating a motion field vector for each player in the plurality of trackable frames. The computing system predicts a future motion of a player based on the player's motion field vector using a neural network.

METHOD FOR GENERATING INSTRUCTION SEQUENCE, ELECTRONIC DEVICE, AND STORAGE MEDIUM

Resumen de: US20260186783A1

0000 Disclosed are a method for generating an instruction sequence, an electronic device, and a storage medium. The method for generating an instruction sequence includes: determining a computation graph corresponding to a neural network model to be compiled and resource status information on hardware executing the instruction sequence; partitioning the computation graph, to determine multiple computation sub-graphs; generating, based on the computation sub-graphs and the resource status information, instruction sub-sequences corresponding respectively to the computation sub-graphs; and determining, based on the instruction sub-sequences, a target instruction sequence corresponding to the neural network model to be compiled.

GENERATING VIDEOS USING SEQUENCES OF GENERATIVE NEURAL NETWORKS

Resumen de: US20260187995A1

0000 Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes receiving a text prompt describing a scene; processing the text prompt using a text encoder neural network to generate a contextual embedding of the text prompt; and processing the contextual embedding using a sequence of generative neural networks to generate a final video depicting the scene.

APPARATUS AND METHOD FOR GENERATING VIDEO DESCRIPTIONS BASED ON ARTIFICIAL NEURAL NETWORKS

Resumen de: US20260188010A1

An apparatus for generating video descriptions according to an embodiment is a video description generating apparatus based on artificial neural networks, including one or more processors and a memory storing one or more programs executed by the one or more processors, and the apparatus includes a feature generating module that extracts a plurality of preset features from a video and generates one synthetic feature based on the plurality of extracted features and a description generating module that receives the synthetic feature and outputs a description text for the video.

HETEROGENEOUS TREE GRAPH NEURAL NETWORK FOR LABEL PREDICTION

Resumen de: US20260187451A1

0000 A method for making predictions includes identifying, for each node in the heterogeneous graph structure, a set of node-target paths that connect the node to a target node; assigning, to each of the node-target paths, a path type identifier indicative of a number of edges and corresponding edge types in the associated node-target path; and extracting a semantic tree from the heterogeneous graph structure. The semantic tree includes the target node as a root node and defines a hierarchy of metapaths that each correspond to a subset of the node-target paths in the heterogeneous graph structure assigned to a same path type identifier. The semantic tree is encoded by generating a metapath embedding corresponding to each metapath in the semantic tree. A label is predicted for the target node in the heterogeneous graph structure based on the set of metapath embeddings.

KNOWLEDGE-AUGMENTED FEATURE ADAPTER FOR VISION-LANGUAGE MODEL NEURAL NETWORKS

Resumen de: US20260187991A1

0000 Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing a multi-modal machine learning task on a network input that includes text and an image to generate a network output. One of the systems includes a vision-language model (VLM) neural network. The VLM neural network includes a VLM backbone neural network and an attention-based feature adapter. The VLM neural network has access to an external dataset that stores multiple text items.

THERMAL IMAGE DEPTH ESTIMATION METHOD AND DEVICE

Resumen de: WO2026141769A1

Disclosed are a thermal image depth estimation method and device. According to the present invention, provided is a thermal image depth estimation device comprising a processor, and a memory stored in the processor, wherein the memory stores program instructions executed by the processor so as to: develop a prototype neural network model by using a pre-collected thermal image dataset; train the neural network model on the basis of a preset loss; and optimize and distribute the trained neural network model according to an open neural network exchange (onnx) format.

DECOUPLED QUERIES FOR END-TO-END 3D TRACKING USING TRANFORMER NEURAL NETWORKS

Resumen de: US20260187138A1

In various examples, a technique for multiple object tracking is disclosed that includes generating, using one or more processing units, one or more first encoded image features based on a first image. The technique also includes generating a plurality of first object embeddings based on the first encoded image features, wherein at least one first object embedding of the plurality of first object embeddings corresponds to a different object depicted in the first image. The technique further includes determining, using a first track query of one or more track queries and one or more machine learning operations, a first association between a first track and a first object that corresponds to the at least one first object embedding, wherein the first track corresponds to the first track query. The technique further includes computing, based on the first association, an object trajectory associating the first track with the first object.

METHOD AND SYSTEM FOR IMPLEMENTING PANEL DEFECT DETECTION ON BASIS OF IMAGE GRAYSCALE EQUALIZATION

Resumen de: WO2026137839A1

The present disclosure relates to the technical field of panel defect detection, and provides a method and system for implementing panel defect detection on the basis of image grayscale equalization. The method comprises: collecting statistics about grayscale values of a panel image by means of image grayscale value statistics collection, so as to obtain a grayscale histogram of the panel image; on the basis of the grayscale histogram of the panel image, using an adaptive histogram equalization algorithm to perform grayscale value equalization processing on the panel image, so as to obtain an equalized image; performing feature labeling on the equalized image, and performing neural network training on the basis of the feature-labeled equalized image, so as to obtain a target detection model; and performing panel defect detection on the equalized image on the basis of the trained target detection model, so as to obtain a defect detection result. By performing grayscale value statistics collection and grayscale value equalization processing on the panel image, defect imaging features can be enhanced, so that the target detection model can accurately perform defect detection and locating, thereby solving the problem that existing Mura defect detection is prone to missed detection.

MULTI-DEVICE COMBINED USE-BASED IDENTIFICATION METHOD AND SYSTEM FOR RAW MATERIALS AND EXCIPIENTS IN FORMULATION

Resumen de: WO2026138708A1

A multi-device combined use-based identification method and system for raw materials and excipients in a formulation. The method comprises: performing cryogenic polishing on a cross-section of a test sample under analysis to obtain a cross-section under test; using an electron microscope to observe the cross-section under test to obtain an electron microscopy image of the cross-section under test, and determining a region under test on the cross-section under test; using an energy dispersive spectrometer to scan the region under test to obtain an energy spectrum image; on the basis of the electron microscopy image and the energy spectrum image, determining whether there are characteristic elements in the test sample, and if there are characteristic elements, identifying the characteristic elements on the basis of the energy spectrum image; and on the basis of the electron microscopy image and the energy spectrum image, determining whether there are components matching preset morphological characteristics in the test sample, and if there are components matching the preset morphological characteristics, using a pre-constructed convolutional neural network model to perform analysis processing on the electron microscopy image to identify the components matching the preset morphological characteristics. The identification method can more accurately and quickly identify components in a sample under analysis.

ANOMALY DETECTION SYSTEM, ANOMALY DETECTION METHOD, ELECTRONIC DEVICE AND STORAGE MEDIUM

Resumen de: US20260187231A1

0000 An anomaly detection system, an anomaly detection method, an electronic device and a storage medium are provided, which relates to the field of artificial intelligence. The anomaly detection system includes a plurality of markers. The plurality of markers are different from each other and correspond to different anomaly detection operations. A log line of a target system may be detected from different dimensions by using the plurality of markers, and marker tokens generated by different markers are obtained. In addition, the anomaly detection system further includes a model marker. The model marker may combine the marker tokens respectively generated by each marker into a token sequence and perform anomaly detection on the token sequence by using a pre-trained neural network model to obtain an overall anomaly score of the target system within a preset time period.

NEURAL NETWORK MODEL TRAINING METHOD AND COMMUNICATION APPARATUS

Resumen de: US20260187461A1

0000 This application provides a neural network model training method and a communication apparatus. The method includes: A transmit-end device performs, by using a first neural network model, forward inference on training data included in a first training process, to obtain a first forward inference result; and sends the first forward inference result to a receive-end device. Further, the transmit-end device receives a first feedback result for the first forward inference result from the receive-end device, and obtains a first intermediate result generated when forward inference is performed on training data included in the first training process. The transmit-end device calculates a gradient of the first training process based on the first feedback result, the first forward inference result, and the first intermediate result. Further, the transmit-end device updates a first model parameter to a second model parameter based on the gradient corresponding to the first training process.

Graph Neural Network-Based Node Classification Method and System, and Related Device

Resumen de: US20260187410A1

A graph neural network-based node classification method, during training of a model, categories of a plurality of neighboring node samples in a graph data sample are first predicted to obtain category distribution of the plurality of neighboring node samples, and sampling is then performed on the plurality of neighboring node samples based on the category distribution and a sampling parameter input by a user to obtain a plurality of sampled nodes, so that category distribution of the plurality of sampled nodes is similar to or consistent with the category distribution of the plurality of neighboring node samples.

MODEL TRAINING METHOD, VEHICLE CONTROL METHOD, AND RELATED APPARATUS

Resumen de: US20260187456A1

0000 Provided are a model training method, a vehicle control method, and a related apparatus, which may be applied to the field of artificial intelligence. The method includes: obtaining road condition information of a target vehicle; obtaining target information based on the road condition information by using a first neural network model, where the target information is a driving intention prediction of the target vehicle, a driving route prediction, or an interaction behavior prediction between the target vehicle and an environment; and updating the first neural network model based on the target information by using an expert system or a result obtained through processing the road condition information and a label corresponding to the road condition information by the expert system.

GENERATIVE AI VIRTUAL CLOTHING TRY-ON

Resumen de: US20260187881A1

0000 An artificial intelligence (AI) network or neural network is trained, using a relatively small number of reference images of a target garment, to enable virtual clothing try-ons of the target garment. Example methods include determining a pose for a person depicted in an input image, determining an area of the input image to replace with a target garment, changing values of pixels within the area, and inputting the pose, the area, and a text prompt describing the target garment, into a neural network, to generate an output image, wherein the neural network is trained to generate the target garment. Example methods include training the neural network with images of clothing in a same class or category as the target garment to teach the neural network to shape the target garment in accordance with a pose of the person and to preserve other clothing and the background.

Compiler and Firmware Mapping Framework for Column-Oriented Neural-Network Compute Arrays

Resumen de: US20260187188A1

0000 A compiler and firmware-generation framework for column-oriented neural-network compute arrays is disclosed. The compiler maps layer operations of a neural network onto a column-based architecture by aligning matrix dimensions to the number of available columns and producing instruction streams that broadcast activation values from each processing row concurrently to all columns while preventing transfer of partial sums between columns. The framework optimizes memory layout, scheduling, and synchronization to preserve the CASCADE dataflow. It can integrate fault-aware remapping and generate microcode for asynchronous or redundant-column configurations. The compiler abstracts the hardware topology, allowing neural-network models to execute efficiently on column-oriented arrays without manual tuning or loss of dataflow integrity.

UNIFIED FEATURE EXTRACTION VIA DYNAMIC FUSION

Nº publicación: US20260187990A1 02/07/2026

Solicitante:

NVIDIA CORP [US]
NVIDIA Corporation

Resumen de: US20260187990A1

System and methods for generating unified feature maps based on an input image. According to one or more embodiments, neural network architectures and machine learning techniques are provided for extracting features from a variety of domains (e.g., geometric, semantic, and auxiliary) and performing dynamic fusion to integrate the extracted features into robust representations, thereby enabling efficient and generalized matching for a variety of computer vision applications, e.g., 3D reconstruction, object tracking, and image retrieval.