190 results about "Image reference" patented technology

The invention relates to the technical field of image segmentation, in particular to an image anaphora segmentation method based on autoregression vertex generation and language structure guidance, and the method comprises the steps: obtaining an input image and a corresponding anaphora query statement; acquiring multi-scale visual features of the input image by using a pre-trained visual encoder; obtaining language features of the anaphora query statement by using a pre-trained language encoder; constructing an image semantic scene graph based on the multi-scale visual features; constructing a query language dependency graph based on the language features; utilizing a graph alignment mechanism and a feature alignment mechanism to realize structured multi-modal fusion of the image semantic scene graph and the query language dependency graph; based on a structured multi-modal fusion result, generating a contour point sequence of the target instance by adopting a language-guided autoregression instance generation method; generating a segmentation mask of the target instance according to the contour point sequence; and outputting the segmentation mask as an image anaphora segmentation result, thereby realizing remarkable performance improvement.

The invention discloses a zero sample anomaly detection method based on multi-scale feature aggregation and semantic guidance, and an abnormal image detection model in the method is obtained through the following steps: obtaining an image training set, an image reference set and text category features; performing multi-scale window feature extraction on the image reference set and the image training set to obtain an image feature set; according to the text category features, performing multi-scale semantic alignment on the first image feature set to obtain semantic guidance prediction representation; performing multi-scale similarity aggregation on the image feature set to obtain an abnormal segmentation prediction representation; and according to the semantic guidance prediction representation and the abnormal segmentation prediction representation, performing parameter updating on the initialized abnormal image detection model to obtain a trained abnormal image detection model. According to the method, the sample data volume required during model training can be effectively reduced, the model training efficiency is improved, meanwhile, the detection precision of the model in a few-sample scene is improved, and the expansibility is high. The invention relates to the technical field of computer vision.

The invention relates to a scene text detection method and device based on a VMama framework and a channel-space attention mechanism and computer equipment. The method comprises the following steps: acquiring a scene image reference sample data set and an enhanced sample data set containing texts; constructing a deformable VMama feature encoder by adopting deformable convolution based on a VMama framework, and obtaining a multi-scale text instance feature map corresponding to each scene image; adopting interpolation and maximum pooling methods to obtain a balance feature map; constructing a GASM module based on a channel-space attention mechanism to obtain text instance features after pixel-by-pixel enhancement; constructing a Fourier feature decoder to obtain a reconstructed text instance contour; and training a scene text detection model which takes the enhanced sample data set as input and takes the text instance contour as output by adopting a loss function. By adopting the method, scene text detection with a global receptive field, rapid convergence and accurate positioning can be realized.

The invention relates to a quality management method and system of a single-port elastic test assembly and a terminal, and relates to the field of product quality detection.The quality management method comprises the steps that image detection information and ultrasonic detection information of a chip contact assembly in a preset detection area are obtained; selecting a chip contact assembly image from the image detection information box based on a preset chip contact assembly feature; determining model information from a preset model database according to the chip contact assembly image; determining fiber core reference information and cladding image reference information according to the model information; determining fiber core detection information according to the ultrasonic detection information; determining whether the fiber core reference information is consistent with the fiber core detection information or not and whether the image detection information is consistent with the cladding image reference information or not; and if the fiber core reference information is inconsistent with the fiber core detection information or the image detection information is inconsistent with the cladding image reference information, repairing by a preset repairing method. Therefore, the quality of the produced and manufactured chip contact assembly is improved.

A method includes incorporating information from multi-modal embeddings in an indexed point-cloud data structure with three-dimensional (3D) spatial information for a captured scene. The method also includes performing at least one of querying or retrieving 3D-point-cloud data based on a user input including at least one of natural language or image references. The method also includes enabling global scene understanding and local scene understanding using instance segmentation in conjunction with the multi-modal embeddings.

The invention aims to disclose an image anaphora segmentation method based on multilevel feature fusion and dual-channel information enhancement. The method comprises the following steps: obtaining and extracting visual features to be segmented and language features in an anaphora text; self-adaptively matching visual features of different levels by using language features, and performing spectrum enhancement on the visual features by using Fourier transform; through cross-modal fusion of visual and language features of different hierarchies, multi-level visual features with language perception are constructed; constructing visual language fusion features by using dual-channel information enhancement; and generating a segmentation mask. According to the method, the advantages of a large model, adaptive dynamic matching and feature enhancement are fused, automatic generation of the segmentation mask is realized, the efficiency and accuracy are improved, and the method has wide application and popularization values.

The invention discloses a feature extraction method and system based on heterogeneous multi-modal information fusion, and the method comprises the steps: constructing a spectrum reference system through synchronously collecting the data of an image sensor, a laser radar and an infrared spectrometer carried by an unmanned plane, achieving the unified calibration of image data, and generating a unified image reference through employing a feature field fusion technology; extracting multi-modal characteristic spectral lines, performing complementary analysis to confirm fusion characteristics, and establishing a dynamic weight table by matching working frequency bands; a separation factor is generated based on correlation analysis of a point cloud density field and a modal characteristic spectral line, a correction coefficient is generated in combination with characteristic fluctuation analysis, and a pure visual characteristic signal is identified; a mapping transformation kernel is constructed to realize standardized conversion of feature data, a migration coefficient is generated through spatial positioning analysis and feature intensity analysis, a regulation and control instruction is generated through multi-level feature processing, a weight table is adjusted based on the regulation and control instruction to form closed-loop optimization, a hierarchical feature extraction signal is generated, and deep fusion and intelligent extraction of features are realized.

The invention discloses an intelligent interactive teaching system and method for oral English emotion based on deep learning, and relates to the field of speech evaluation. The system comprises an environment detection and evaluation module, a threshold comparison module, an audio and video collection and processing module, an optimization module and a suggestion output module. Threshold value range comparison and image reference threshold value comparison are carried out on the spoken English intelligent interaction environment evaluation coefficient obtained in the environment detection and evaluation module, and threshold value comparison is carried out on the explosion loss judgment evaluation coefficient of the spoken English intelligent interaction audio obtained in the audio and video collection and processing module. According to the method, the intelligent oral English interaction audio data input by the user are evaluated to obtain the explosion judgment evaluation coefficient of the intelligent oral English interaction audio, and then the text, the audio and the visual oral English error correction suggestion are output according to the explosion judgment evaluation coefficient of the intelligent oral English interaction audio, so that the problem that in the prior art, the user experience is poor is solved. And the accuracy of judging the explosion in the context collaborative pronunciation change is insufficient.

The invention discloses a digital role audio and video generation method and a digital role live broadcast interaction method, and the generation method comprises the steps: generating explanation data of a to-be-explained object according to a digital role material generation task of a requester; performing data conversion processing on the explanation data to obtain audio data corresponding to the explanation data; performing role image digital processing on the obtained role image reference data to obtain a digital role image video; carrying out fusion processing on the audio data and the digital role image video to obtain a digital role driving video for explaining the object to be explained by using a digital role; returning the digital role driving video to the requester, so that the requester processes the digital role driving video before distribution to generate a digital role audio / video; therefore, computing resources are saved, the cost is reduced, and the coverage rate of digital role audio and video playing is improved.

Systems and methods are described herein for processing video. An encoder may generate, for a sequence of video frames, a plurality of first segments and a plurality of second segments. The plurality of first segments may comprise stream access points (SAPs) of a first type that do not reset a picture reference buffer. The plurality of second segments may comprise SAPs of a second type that do reset the picture reference buffer. The encoder may send segments of the plurality of first segments to a computing device streaming video when network conditions are steady. The encoder may send a segment of the plurality of second segments following a switch, by the computing device, to a different bitrate based on a change to the network conditions. Once the computing device has decoded the segment the plurality of second segments, the encoder may send subsequent first segments at the different bitrate.

The embodiment of the invention provides an image generation method, and the method comprises the steps: obtaining a user portrait of a target user based on an image generation service request initiated by the target user; performing matching processing on the user portrait and a preset image style template library through a user style portrait prediction model to obtain a user style portrait prediction result of the target user; generating an image generation template drawing of the target user according to the user style portrait prediction result and the preset image style template library through an image generation model; and through the image generation model, generating a target image of the target user according to the image generation template image and the target image reference image. Through the embodiment of the invention, the problem that a personalized image conforming to the preferences and features of the user cannot be automatically generated according to the preferences and features of the user, the face features and the like in the related technology can be solved, so that the generated target image is more accurate and better fits the preferences and features of the target user.

The invention provides a video reference frame management method applying a real-time scene. The method comprises the following steps: updating a reference frame queue: (1) inputting an image Fn to be compressed; (2) generating a reconstructed image F'n from the to-be-compressed image Fn through a coding inverse process; (3) sending the reconstructed image F'n into an image reference frame queue and a traditional reference frame queue at the same time for respective update processing; comprising the following steps of: (1) inputting an image Fn to be compressed; (2) generating a scene ID of the image; (3) comparing the scene ID of the generated image with the scene ID of the scene reference frame queue; (4) judging whether the scene ID of the to-be-compressed image Fn is the same as the scene ID in the scene reference frame queue or not; (5) if yes, taking the scene reference frame images with the same scene ID as reference frames of the to-be-compressed image; and (6) if not, obtaining the reference frame from the traditional reference frame queue. According to the method, the video compression efficiency can be remarkably improved.

The invention belongs to the technical field of computer vision, and discloses an interactive semi-automatic anomaly detection labeling method, system and device and a medium, and the method comprises the steps: obtaining a to-be-detected image of an industrial product, manual click input representing an abnormal region position, and natural language description of a defect type; based on the defect-free image reference set, extracting position constraint features of each image patch of the to-be-detected image, and calculating pixel-level position rapid abnormal residual features; converting the natural language description into a discriminant language feature vector by using a pre-training language model and a discriminant encoder; processing the residual feature through a residual feature branch, and generating a first fusion feature through a cross-modal attention mechanism under the guidance of a language feature vector; processing the original pixel information by combining the pixel information branch with click input to generate a second fusion feature; and fusing the two features and decoding to obtain a pixel-level abnormal annotation mask, thereby realizing high-precision semi-automatic annotation of the abnormal region in the industrial product image.

Provided is an inspection device capable of obtaining accurate coordinates of a chip provided on a substrate. Specifically, this inspection device comprises: a stage on which a target substrate is placed; a camera that images the target substrate to obtain a captured image; and a controller that calculates a target measurement value of target coordinates of a chip with respect to a target origin of the target substrate on the basis of the distance between an image reference position and the chip in the direction in which pixels are arranged and a relative movement set value between the stage and the camera. The controller corrects the target measurement value by using a projective transformation matrix to calculate a target correction value. The controller calculates an alignment measurement value of the alignment coordinates with respect to an alignment origin of an alignment substrate on the basis of the distance between the image reference position and an alignment mark in the direction in which the pixels are arranged, in a state in which the alignment origin is aligned with the image reference position. The controller calculates a parameter of the projective transformation matrix on the basis of the alignment measurement value and a known alignment actual measurement value.

The invention relates to a remote sensing image anaphora segmentation method and system based on multi-mode prompt, and belongs to the technical field of computer vision and remote sensing intelligent interpretation. The method comprises the following steps: S1, data preprocessing: dividing a remote sensing image anaphora segmentation data set; s2, constructing a remote sensing image anaphora segmentation model based on multi-modal prompt, wherein the remote sensing image anaphora segmentation model comprises an image encoder, a space prior module, a text encoder, a fine-grained visual language interaction device and a multi-modal prompt encoder; s3, training a remote sensing image anaphora segmentation model based on multi-modal prompt: for the training of the remote sensing image anaphora segmentation model, performing optimization training on the model by using a loss function combining Focal loss, cross entropy loss and Dice loss; and S4, outputting a segmentation result: inputting a remote sensing image to be segmented into the trained model, and then inputting a model output result into a mask decoder to complete anaphora segmentation of the target object. According to the invention, the interpretation efficiency is improved.

The invention relates to the field of image data processing, and provides a steam boiler gas leakage detection method based on image processing, which comprises the following steps: setting an image acquisition unit, covering a predetermined leakage risk area together, acquiring corresponding reference images under different production working conditions, and establishing an image reference library covering a target area; the method comprises the following steps: dividing a boiler operation time axis into a plurality of production states, calling corresponding reference images according to different states, and carrying out difference, partition comparison and motion detection on the real-time acquired images to obtain suspected leakage cloud cluster areas in the corresponding states; arranging a condensate water indication patch, collecting a patch image, extracting through color, brightness and texture features to form a patch humidity change curve, carrying out multi-source information fusion on the patch humidity change curve, time sequence features and production time states, and outputting a corresponding steam leakage risk level; and generating alarm information according to the steam leakage risk level, and recording and archiving the alarm information in a monitoring and management system for energy consumption analysis and risk tracing.

The invention relates to an image editing method for rotating a single target in an image based on a diffusion model, and belongs to the field of image editing. According to the method, an image editing network based on a diffusion model is provided; adding a visual angle picture generated by an image-3D technology into the editing network as a reference picture; the noise sampled by the diffusion model is optimized through the comparison loss of the edited picture, the original picture and the reference picture; and finally, the target rotation effect is achieved. In order to keep the consistency of background contents, the region-based image filling is added into the sampling process of the diffusion model; in addition, super-resolution processing is carried out on the visual angle picture generated by the image-3D technology, and the editing quality is further improved. By means of the method, good performance can be effectively achieved on the object rotation editing task.

The invention relates to a 3D positioning method and system based on a holder, and the method comprises the following steps: obtaining an original image according to a camera connected with the holder, and obtaining a theoretical rotation angle of the holder according to the original image; controlling the holder to rotate according to the theoretical rotation angle, so that the central point of the region of interest is located in the center of the image to obtain a 3D positioning image; respectively extracting image feature points of the original image and the 3D positioning image, and determining an original image reference point and a 3D positioning image reference point based on the feature descriptors of the original image and the 3D positioning image so as to complete image feature point matching; and performing 3D positioning error correction according to the deviation of the original image reference point and the 3D positioning image reference point to complete 3D positioning. According to the method, the reference point is determined by extracting the geometric feature points of the image and generating and matching the descriptors, and finally 3D positioning error correction is performed through the deviation of the reference point, so that 3D positioning is completed, and the precision of 3D positioning can be greatly improved.

To provide an inspection apparatus with which accurate coordinates of a chip provided on a board can be obtained.SOLUTION: An inspection apparatus includes: a stage on which a target board is placed; a camera for imaging the target board to obtain a captured image; and a controller which calculates a target measurement value of target coordinates of a chip with respect to a target origin of the target board, on the basis of a distance between an image reference position and the chip in a pixel arrangement direction and a relative movement set value between the stage and the camera. The controller corrects the target measurement value by a projective transformation matrix to calculate a target correction value. The controller calculates an alignment measurement value of alignment coordinates with respect to an alignment origin, on the basis of a distance between the image reference position and an alignment mark which is aligned with the alignment origin of an alignment board. The controller calculates a parameter of the projective transformation matrix on the basis of the alignment measurement value and a known actual alignment measurement value.SELECTED DRAWING: Figure 1

The invention relates to an underwater image enhancement method, system and device combined with attenuation channel correction, and the method comprises the following steps: carrying out the color correction of an image through employing an attenuation channel adaptive correction white balance technology, and recording the image as a first enhancement image I1; de-noising the first enhanced image I1 by using a fast non-local mean algorithm based on a local block, and recording the de-noised first enhanced image I1 as a second enhanced image I2; improving the contrast ratio of the first enhanced image I1 by adopting a self-adaptive stretching method based on the local characteristics of a histogram, and recording the first enhanced image I1 as a third enhanced image I3; and obtaining a final enhanced image by using a bilateral weight fusion strategy in combination with the complementary advantages of the second enhanced image I2 and the third enhanced image I3. The system and the equipment are used for executing the method. According to the method, qualitative and quantitative experiment results on different underwater image reference data sets show that the contrast of the image can be uniformly improved and details and textures of the image can be enhanced while the color of the image is corrected.

The invention discloses an RAG document splitting optimization method and system in the operator field, and belongs to the technical field of large model optimizing.The method comprises the steps that a document is uploaded, and source files and file information are stored through minio and paradeDB respectively; constructing a document loader, automatically selecting a corresponding loader according to a file type to carry out file processing, converting the file into a uniform Markdown text, analyzing a picture in the document, and converting the picture into a base64 format; constructing a picture processor, extracting a picture base64 character string in the text, processing a bitmap and a vector diagram, and converting the bitmap and the vector diagram into a Markdown picture reference format; constructing a document divider; vector conversion; and recalling the text. According to the method, the problem that picture extraction, conversion and processing are complex in the current RAG is solved, vector retrieval, full-text retrieval and mixed retrieval functions can be achieved through one database, and the operation and maintenance cost is reduced.

The embodiment of the invention provides a video image component prediction method and device, and a computer storage medium. The method can comprise the following steps: acquiring a reference value set of a first image component of a current block; determining a plurality of first image component reference values from a reference value set of the first image component; performing first filtering processing on the sample values of the pixel points corresponding to the plurality of first image component reference values to obtain a plurality of filtered first image reference sample values; determining to-be-predicted image component reference values corresponding to the plurality of filtered first image reference sample values; determining parameters of a component linear model according to the plurality of filtered first image reference sample values and the to-be-predicted image component reference value; according to the component linear model, mapping the reconstruction value of the first image component of the current block to obtain a mapping value; and determining a predicted value of the to-be-predicted image component of the current block according to the mapping value.

This application provides a video image component prediction method and apparatus, and a computer storage medium. The method may include: acquiring a reference value set of a first image component of the current block; determining multiple first image component reference values ​​from the reference value set of the first image component; performing a first filtering process on the sample values ​​of the pixels corresponding to the multiple first image component reference values ​​respectively to obtain multiple filtered first image reference sample values; determining the image component reference value to be predicted corresponding to the multiple filtered first image reference sample values; determining the parameters of a component linear model based on the multiple filtered first image reference sample values ​​and the image component reference value to be predicted; performing a mapping process on the reconstructed value of the first image component of the current block according to the component linear model to obtain a mapping value; and determining the predicted value of the image component to be predicted of the current block based on the mapping value.

Techniques for video processing including video coding, video decoding, and video transcoding are described. One example method includes performing a conversion between a video comprising a video picture and a bitstream of the video, the video picture comprising one or more slices. The video picture refers to a picture parameter set, and the picture parameter set conforms to a format rule that specifies that the picture parameter set includes a column width list of N slice columns, where N is an integer. The (N-1) th slice column is present in the video picture, and the width of the (N-1) th slice column is equal to the (N-1) th entry in the explicitly included slice column width list plus a number of codec tree blocks.

The invention discloses a video generation method and device, a model training method and device and a computer program product, and the video generation method comprises the steps: obtaining a target audio and a reference picture used for generating a video, and the reference picture comprises a sound production object; determining a global visual feature of each to-be-generated video frame corresponding to the audio clip according to the clip features of one or more audio clips corresponding to the target audio and the reference image; according to the pronunciation feature of each audio frame of the target audio and the lip feature of the sounding object in the reference picture, determining the lip feature of the sounding object in a to-be-generated video frame corresponding to the audio frame; and generating each video frame according to the lip feature and the global visual feature corresponding to the to-be-generated video frame. Through the scheme provided by the invention, the expression of characters in the generated video is more vivid and natural, the lip action and the audio can be accurately synchronized, and the visual experience of a user is improved.

The invention provides an unmanned aerial vehicle image anaphora detection method based on a mixed granularity expert, and belongs to the field of unmanned aerial vehicle image anaphora detection. The cross-modal sequence is input into multiple layers of stacked coding blocks, each layer of coding block comprises an MHSA layer, an MoG layer and an FFN layer which are connected in sequence, and after multi-layer iteration, a cross-modal scale comprehensive memory is output; the cross-modal scale comprehensive memory and the target query are input into a multi-layer stacked scale comprehensive decoder, each layer of decoder comprises an MHSA layer, an MHCA layer, an MoG layer and an FFN layer which are connected in sequence, and coarse query representation is output; the coarse query representation and the cross-modal scale comprehensive memory are input into a multi-layer stacked scale sensitive decoder, each layer of decoder comprises an MHSA layer, an MMCA layer and an FFN layer which are connected in sequence, after multi-layer iteration, the output fine-grained query representation enters a regression prediction head, and an anaphora detection result is obtained; the invention further provides a detection system. The problem that an existing method is low in detection precision in an aerial photography scene is solved.

The invention relates to an image generation method and device, electronic equipment and a storage medium. The method comprises the steps that a preset page comprising at least one kind of image reference information is displayed; in response to a reference image selection operation for first image reference information, determining at least one reference image corresponding to the first image reference information, the first image reference information being any one piece of image reference information in the at least one piece of image reference information; in response to the image generation operation, displaying at least one generated image, the at least one generated image being generated based on target reference information corresponding to each of a plurality of target reference images, the plurality of target reference images being reference images determined when the image generation operation is triggered, the plurality of target reference images correspond to one or more pieces of image reference information in the at least one piece of image reference information. According to the embodiment of the invention, a user can flexibly and conveniently select the image reference information and the corresponding reference image, so that the image generation effect is improved.

The invention relates to an automatic generation method, system and device for a clinical image reference map and a medium, and the method comprises the steps: a doctor determines the anatomical positioning of a required reference map according to a browsed image, and extracts the image examination category, examination items and basic information of a patient of the patient, and the relative positions of the layer where the anatomical positioning is located in all scanning layers; the method comprises the following steps: matching same image examination items and categories in a hospital image diagnosis report database, matching basic information of a patient, examination equipment and the like according to a priority set by a doctor, determining a retrieval keyword according to anatomical positioning, determining a retrieval structured medical image report statement based on the retrieval keyword, and querying the hospital image report database. Matching the report statement and judging the report as a normal report; and calling the report image, matching a reference atlas layer based on the relative position of the anatomical positioning layer in all scanning layers, automatically generating a reference atlas, and displaying the reference atlas to a physician for reference. According to the invention, accurate and abundant imaging reference images can be provided for imaging diagnosis doctors, and the working efficiency can be improved.

Arrangements are disclosed that enable reliable and accurate referencing of image distance, orientation, and / or color. According to exemplary embodiments, an image reference marker is arranged on or near an area to be imaged. A display shows a live view of the area with a superimposed alignment pattern, while a user positions the image capture device so that the image of the marker is aligned with the alignment pattern. Once satisfactory alignment is achieved, an image of the area can be captured manually or automatically.

Arrangements are disclosed which provide reliable and accurate image ranging, orientation, and / or color referencing. In exemplary implementations, an image reference marker is arranged at or proximate to an area to be imaged. A display shows a live view of the area with a matching pattern superimposed thereon as a user positions the image capture device to align the image of the marker with the matching pattern. Once satisfactory alignment is achieved, an image of the area can be captured manually or automatically.