122 results about "Diffusion network" patented technology

The invention discloses an industrial part defect sample accurate generation method based on a conditional diffusion model, and belongs to the field of image processing and artificial intelligence. The method forms a closed-loop cooperative system by constructing four deep coupling modules of physical constraint noise scheduling, multi-scale feature coupling, double-domain feedback optimization and adaptive weight adjustment; a defect physical forming mechanism is converted into a dynamic noise scheduling strategy, deep interaction between condition information and a feature map is established at multiple levels of a diffusion network, quality closed-loop optimization is achieved through dual evaluation of a pixel domain and a frequency domain, and training weight is dynamically adjusted according to defect scarcity. And multi-scale accurate control is realized, a quality guarantee closed loop is established, the problem of data imbalance is effectively solved, and the performance of an industrial defect detection model is remarkably improved.

Approaches presented herein provide for the generation of synthesized data from input noise using a denoising diffusion network. A higher order differential equation solver can be used for the denoising process, with one or more higher-order terms being distilled into one or more separate efficient neural networks. A separate, efficient neural network can be called together with a primary denoising model at inference time without significant loss in sampling efficiency. The separate neural network can provide information about the curvature (or other higher-order term) of the differential equation, representing a denoising trajectory, that can be used by the primary diffusion network to denoise the image using fewer denoising iterations.

The invention provides an unbiased missing modal learning method based on a multi-stage double diffusion network, and belongs to the field of computer vision, natural language processing and multi-modal information fusion. The method comprises the following steps: a multi-modal feature extraction module maps text modal data, image modal data and audio modal data into a unified potential representation space; the potential space representation of the three modals is used as the original feature of the missing modality and the original feature of the available modality; based on the missing modal original features and the available modal original features, training a multi-stage double-diffusion module through forward diffusion and reverse diffusion to obtain a trained multi-stage double-diffusion module, and based on the trained multi-stage double-diffusion module, performing global structure generation and modal conversion through available modal data to obtain a multi-stage double-diffusion model; and performing local detail optimization reasoning to generate missing modal data. According to the method, the problem of modal generation deviation in the existing multi-modal learning is solved, and the effectiveness of the multi-modal learning effect in the missing scene is enhanced.

The invention provides a method for predicting a cabin material pile in a ship loading process of a ship loader based on a diffusion network, and relates to the technical field of bulk cargo loading and unloading automation. Three-dimensional point cloud data in a cabin in the ship loading process is acquired, the aligned three-dimensional point cloud data is projected to a two-dimensional plane parallel to the bottom of the cabin, and a material pile depth map is generated; constructing a training sample pair based on the historical ship loading operation data, and training a conditional diffusion network model to obtain a target conditional diffusion network model; inputting the material pile depth map of the first time period and planned blanking point information corresponding to the second time period into a target condition diffusion network model to obtain a material pile depth map corresponding to the second time period; and taking the material pile depth map corresponding to the second time period as an initial depth map for prediction in the next time period to carry out iterative prediction, and outputting a material pile depth prediction set after the material pile depth map prediction in all time periods in the ship loading process is completed. According to the invention, the fidelity and authenticity of the predicted material form can be improved.

The invention discloses a background coherent story picture book generation method based on a diffusion model, and belongs to the field of computer vision and generative artificial intelligence. The method comprises a training stage and a testing stage: in the training stage, bidirectional cross attention fusion and modal soft selection are carried out on text, background and role multi-modal conditions through a feature enhancement fusion module, model optimization is carried out by utilizing joint alignment loss, and efficient parameter fine tuning is carried out on a diffusion model by adopting an efficient parameter fine tuning method; in the test stage, a reference image input by a user and a text sequence are processed into a fusion condition, a large language model is driven to generate an image mark, the image mark is converted into a diffusion condition through a mapper, and each frame of image is generated step by step by combining an autoregression mode with a multi-condition injection diffusion network. According to the method, the problem of inconsistency of cross-frame backgrounds, styles and roles is effectively solved, and high-quality and coherent generation of the long-sequence story picture book is realized.

A computer-implemented system for foreign object detection in a scene. The system includes a memory-suppress diffusion network module adapted to reconstruct a reconstructed image from an encoded image, and a contrastive dissimilarity network adapted to combine the input image and the reconstructed image to predict an anomaly map for the input image. The encoded image is based on an input image, and the memory-suppress diffusion network module and the contrastive dissimilarity network are trained using only normal, real images. The system leverages only normal images in training and does not compromise the detection performance at the inference stage.

The invention belongs to the technical field of computers, and provides a video virtual fitting method and device based on a diffusion model and a program product. The method comprises the following steps: firstly, acquiring a target clothing graph and a model video, and extracting depth features, masks and a model latent feature video; meanwhile, semantic codes of the clothes are extracted through a CLIP encoder. The first diffusion model processes the clothing graph and the CLIP code, and outputs a clothing space attention feature vector fusing fine-grained texture and semantics; and the second diffusion model serves as a video diffusion network, receives random noise, character features, masks and depth features, injects garment vectors and CLIP codes into the middle layer, and finally outputs a fitting video with naturally fitted and dynamically coherent garments. According to the method, the high-fidelity virtual fitting video can be efficiently generated according to the new clothing and model video, and the method has wide application value in the fields of electronic commerce and the like.

Approaches presented herein provide for the generation of synthesized data from input noise using a denoising diffusion network. A higher order differential equation solver can be used for the denoising process, with one or more higher-order terms being distilled into one or more separate efficient neural networks. A separate, efficient neural network can be called together with a primary denoising model at inference time without significant loss in sampling efficiency. The separate neural network can provide information about the curvature (or other higher-order term) of the differential equation, representing a denoising trajectory, that can be used by the primary diffusion network to denoise the image using fewer denoising iterations.

The invention discloses a human body posture estimation result generation method and device based on a generative model, relates to the technical field of image processing, effectively recovers a complete skeleton structure, realizes accurate recognition of human body postures in high-shielding and high-dynamic change scenes such as an electric power production field, and improves the adaptability to shielding scenes. The method comprises the following steps: acquiring a to-be-identified video stream, and performing initial attitude estimation on each frame of video image in the to-be-identified video stream by using a pre-trained attitude detection network model to obtain an initial skeleton point sequence; fitting the initial skeleton point sequence by adopting a Gaussian mixture modeling method to obtain a defect skeleton point sequence; inputting the defective skeleton point sequence into a pre-trained inverse diffusion network model, and optimizing the defective skeleton point sequence in combination with the adjacent frame skeleton point coding information of each frame of video image in the inverse diffusion network model to obtain a complete skeleton point sequence; and generating and outputting a human body posture estimation result based on the complete skeleton point sequence.

Provided is a target model training method, a multimodal data processing method, and devices therefor, relating to the field of artificial intelligence technology, and in particular to the fields of computer vision, deep learning, large model and other technologies. The target model training method includes: inputting sample data into a preset model to obtain initial multimodal features of the sample data; and using the initial multimodal features and a preset noise feature to perform model training on N diffusion networks in the preset model to obtain a target model when parameters of an image-text encoder of the preset model are fixed.

The invention discloses an infrared small target detection method based on visual thermal diffusion and selective multi-scale feature fusion, and mainly solves the problems of easy feature loss and low detection precision caused by weak target and complex background in the existing infrared small target detection technology. The implementation scheme is as follows: 1) acquiring a data set and a segmentation label; 2) constructing an infrared small target detection model; 3) constructing a loss function; 4) training an infrared small target detection model; and 5) obtaining an infrared small target detection result. According to the infrared small target detection model constructed by the invention, through a diffusion network encoder pre-trained based on a visual thermal diffusion principle, feature representation of a weak target is enhanced from the source; intelligent fusion and enhancement of multi-level features of the encoder are realized through a selective dimension fusion module deployed on jump connection; and through a selective cavity channel refining module and a selective receptive field fusion module which are embedded in the network, the high efficiency of the whole architecture is ensured while the fine extraction capability of the network to small targets is improved.

The application discloses a medical image segmentation method based on a pre-training diffusion model, which comprises the following steps: 1, obtaining a medical image segmentation dataset and preprocessing; 2, establishing a forward noise adding, reverse noise removing process and a pre-training prediction network; and 3, training a segmentation network and a prediction. The application directly predicts an original label and prior knowledge of a large model through a diffusion network, and improves the segmentation performance of a target region in a medical image.

The application belongs to the technical field of computers, and provides a video virtual try-on method, device and program product based on a diffusion model. First, a target garment image and a model video are acquired, and a deep feature, a mask and a model latent feature video are extracted therefrom; meanwhile, a semantic code of the garment is extracted through a CLIP encoder. A first diffusion model processes the garment image and the CLIP code, and outputs a garment spatial attention feature vector that fuses fine-grained texture and semantics; a second diffusion model serves as a video diffusion network, receives random noise, a person feature, a mask and a deep feature, and injects a garment vector and the CLIP code into an intermediate layer, and finally outputs a try-on video that is natural and dynamically coherent. The application can efficiently generate a high-fidelity virtual try-on video according to a new garment and a model video, and has wide application value in the field of e-commerce.

This invention discloses an underwater image enhancement method based on a diffusion model with red-dark channel prior constraints. The method includes inputting a single-frame underwater raw image into the model and passing it through a joint estimation module of red-dark channel prior and quadtree, outputting transmittance and background light. These are then encoded by a physical condition encoder to generate a physical condition vector containing spatial attenuation priors and global background light constraints. This vector is used by the diffusion network for encoding and utilizes transmittance distribution and global background light information. In the reverse denoising process, a physical consistency layer and the Jaffe-McGlamery equation are combined to construct a physical consistency residual. The residual gradient calculated based on the physical model is injected as a guiding term into the noise prediction process. The enhanced image is iteratively generated through a denoising diffusion implicit model sampling process. The enhanced image is then forward-degraded and re-rendered through a cyclic consistency reconstruction branch. The cyclic consistency loss is calculated and used for reverse optimization of network parameters. This invention can reduce the color shift problems caused by background light misjudgment and transmittance overestimation.

The application discloses a concrete dam crack segmentation method and system based on a diffusion model, which comprises the following steps: 1) preprocessing a water conservancy dam image; 2) encoding a concrete dam crack feature; 3) removing image noise to obtain a crack boundary feature by using a diffusion network; 4) adjusting a predicted crack boundary feature in combination with a known region; and 5) post-processing a dam image crack segmentation result. The application uses a crack boundary feature extraction module to obtain basic crack edge information, uses a Trimap form to guide the training direction of the model, makes the model faster to fit to a global optimal point, then uses a diffusion crack boundary feature restoration module to obtain a refined segmentation result from the Trimap, and finally uses a crack boundary prediction correction module to combine the labeled information, so that the model pays more attention to the foreground crack region rather than the background region, a more accurate feature representation is obtained, the detection precision is improved, the missed detection situation is reduced, and the overall performance of the model is improved.

Devices, systems, and techniques for denoising inference or training data of neural networks are described. In at least one embodiment, the denoising of inference or training data of neural networks can be performed based at least partially on identifying different types of inference or training data within the inference or training data of neural networks, which are to be denoised separately using a corresponding number of neural diffusion networks.

The invention relates to a stochastic dynamics system energy prediction method and device, electronic equipment and a medium. The method comprises the following steps: setting a displacement variable of the stochastic dynamics system about time, and establishing an energy function of the stochastic dynamics system according to the displacement variable; constructing a multi-scale data set according to the energy function; the initial model is trained through the multi-scale data set, and a feature extraction model is obtained after training is completed; the feature extraction model is used for extracting dynamic features of the stochastic dynamic system; the dynamic characteristics at least comprise an energy predicted value at the current moment; constructing a drift term according to the drift network and the dynamic characteristics, and constructing a diffusion term according to the diffusion network and the dynamic characteristics; and calculating an energy predicted value of the next moment based on the energy predicted value of the current moment, the drift term and the diffusion term. According to the method, based on the improved discretization incremental energy prediction algorithm, the discretization scheme conforming to the physical law is adopted to improve the energy prediction numerical stability and prediction precision.

This invention relates to artificial intelligence, and particularly to a supply chain demand forecasting method based on condition-guided temporal diffusion and multimodal fusion. The method includes: acquiring multimodal business samples containing historical sales data with missing masks and graphic descriptions; obtaining cross-modal aligned feature vectors through feature extraction and spatial projection to construct a static multimodal conditional representation; under this guidance, performing inverse denoising sampling using a temporal diffusion network to obtain reconstructed temporal features; generating a measure signal to fill uncertainty through multiple sampling variance calculations, inputting it into a dynamic gating network to adaptively allocate fusion weights, and obtaining global cross-modal attribute features; finally, generating demand forecasting results and performing end-to-end training based on a joint loss function of prediction error and diffusion noise prediction loss. This invention effectively overcomes the problems of supply chain cold start and stockout gaps, significantly improving prediction accuracy and model robustness in extremely sparse scenarios.

The invention discloses a network flow generating and repairing method based on protocol constraint. The method comprises the following steps: analyzing an original traffic capture file, extracting basic features through stream-level recombination and session segmentation, and mapping a continuous time sequence and a discrete protocol field to a uniform feature space; and on the basis, constructing a diffusion type network flow generation model to capture a statistical attribute and time sequence dependency relationship of the network flow. In the generating or repairing process, a protocol state machine is constructed, protocol constraint is introduced in the reverse denoising stage, the sampling process is constrained through a legality guiding mechanism, and protocol logic violation in the generating process is avoided. And finally, through feature inverse mapping and virtual protocol stack state maintenance, reconstructing to obtain network flow data which meets a communication protocol specification and can be correctly analyzed by a real protocol stack. According to the method, the generation and repair of the network traffic can be realized under a unified framework, and the correctness of a result in protocol semantics is ensured.

This invention discloses a method, system, and apparatus for visual image reconstruction based on electroencephalogram (EEG) signals, belonging to the field of visual image reconstruction technology. The method includes acquiring the EEG of imagined images; obtaining evoked scalp EEG signals for each individual image through data preprocessing; inputting the EEG signals of each image stimulus and its corresponding imagined image into a diffusion network encoded with multi-granularity audio templates for training, enabling the network to generate high-quality reconstructed images; and in the testing phase, directly inputting the preprocessed EEG signals of unknown imagined images into the multi-granularity audio template-encoded diffusion network to generate corresponding reconstructed images. This method significantly broadens the scope of visual information reconstruction using scalp EEG signals, fills the gap in the field of scalp EEG signal reconstruction of imagined images, and enables in-depth exploration of brain thought reconstruction.

The present disclosure relates to systems, non-transitory, computer-readable media, and methods for the three-dimensional rotation of vector graphics. In particular, in some embodiments, the disclosed systems provide a two-dimensional vector graphic in a first orientation for display via a graphical user interface of a client device. Additionally, in some embodiments, the disclosed systems receive user input to rotate the two-dimensional vector graphic in a three-dimensional space to a second orientation. Furthermore, in some embodiments, the disclosed systems generate a new two-dimensional graphic using a neural diffusion network, which represents the two-dimensional vector graphic rotated according to the user input. Finally, in some embodiments, the disclosed systems provide the new two-dimensional graphic in the second orientation for display via the graphical user interface.

The invention relates to a stochastic dynamics system energy evolution solving method and device and electronic equipment. The method comprises the following steps: acquiring input data of a current time step; constructing an energy evolution model of the stochastic dynamics system through a stochastic differential equation according to the preset drift network and the preset diffusion network; predicting an energy increment through an energy evolution model according to a preset drift network, a preset diffusion network, the input data and a preset step length, and updating the total energy according to the energy increment; and updating the preset step length based on the energy increment, updating from the current time step to the next time step according to the updated preset step length, and returning to the step of predicting the energy increment through the energy evolution model according to the preset drift network, the preset diffusion network, the input data and the preset step length until the preset termination time is reached. And obtaining an energy evolution trajectory of the stochastic dynamics system. According to the method, the stochastic differential equation of stochastic dynamics system energy evolution can be efficiently and stably solved, and energy numerical simulation precision and calculation efficiency are improved.

The invention discloses an intelligent reflecting surface auxiliary environment point cloud reconstruction method based on a diffusion model. The method is used for solving the problems of point cloud missing and insufficient reconstruction precision caused by shielding in a complex environment. The method comprises the following steps: optimizing the IRS phase of the intelligent reflecting surface so as to enhance the quality of a received signal and improve the multi-view observation capability; based on the optimized signal, utilizing a multi-head self-attention mechanism to extract deep features, and generating an initial coarse point cloud; and then, iterative denoising and diffusion completion are carried out on the incomplete initial point cloud through a point cloud diffusion network, a shielded region is accurately reconstructed, and finally, a high-integrity and high-density three-dimensional point cloud is output. According to the method, high-precision and high-integrity three-dimensional scene reconstruction in a complex shielding environment is realized by deeply fusing a multi-view signal provided by an intelligent reflecting surface and the generation capability of a diffusion model.

A bridge RC pile reinforcement design method and system based on a two-stage diffusion model of mask information constraint, comprising: obtaining the RC pile structure reinforcement design requirements to be processed; extracting key information from the design requirements and performing matrix processing to generate a pile foundation design mask matrix; sampling from random Gaussian noise to obtain an initial noise tensor, inputting the pile foundation design mask matrix and the initial noise tensor into a pre-trained RC pile reinforcement diffusion network model constrained by mask information, and repeatedly executing the step of the pre-trained diffusion network model until the iteration time step reaches a preset value to obtain an RC pile reinforcement design latent feature tensor; inputting the RC pile reinforcement design latent feature tensor in the low-dimensional space into a pre-trained variational autoencoder to obtain an RC pile steel reinforcement arrangement design drawing in the pixel space. The present application realizes efficient and reliable intelligent pile structure reinforcement design and belongs to the field of civil structure engineering and computer deep learning application technology.

The invention relates to the technical field of information technology and intelligent analysis, in particular to a policy diffusion network intelligent simulation analysis system which comprises a subject behavior modeling module, a propagation path analysis module, a social network embedding module and a dynamic environment adaptation module. Through combination of multi-subject behavior modeling and social network embedding, node influence is quantified, an instruction execution time sequence is dynamically recombined, and the simulation optimization problem in a complex policy environment is solved. According to the method, the precision and applicability of the policy diffusion process can be improved, collaborative optimization of the environment variable recovery speed, policy adjustment offset suppression and external disturbance control is realized, and diversified policy research requirements are met.

The invention provides a physical simulation prior and text dual-condition-based nondestructive testing ultrasonic imaging data generation and enhancement method, which comprises the following steps of: firstly, acquiring a training data triple consisting of a real ultrasonic image, a physical prior image and a text label; and constructing a conditional diffusion model comprising a diffusion type generation network, a text encoder and a physical prior encoder. Text tags are encoded into semantic feature vectors, physical prior images are encoded into multi-scale spatial feature maps, and the multi-scale spatial feature maps are jointly injected into a diffusion network, so that double constraints on spatial structures and semantic contents of generated images are realized. In the training stage, a high-fidelity composite image is obtained. According to the method, by designing a multi-scale spatial feature injection mechanism, acoustic structure features of a physical field are deeply coupled with semantic features of a text, strong physical constraint is performed on a diffusion process in a submerged space, and the problems of scarcity and difficult acquisition of real defect samples in an industrial scene are effectively solved; and high-fidelity data coverage from a small sample to a full scene is realized.

The application discloses a CTA image generation method based on a state space diffusion network, and relates to the field of medical image processing. The method uses a state space diffusion network to generate a CTA image according to a non-enhanced CT image. The state space diffusion network introduces an improved state space model in a diffusion generation process to enhance the modeling capability of long-range dependent information, thereby improving the continuity and topological consistency of a blood vessel structure. Meanwhile, an adversarial learning mechanism is combined in the generation framework, and the image detail expression capability is strengthened through a local region discrimination strategy, so that the generated CTA image is closer to a real CTA image in terms of boundary definition and texture restoration. The method can realize high-quality conversion from a non-enhanced CT image to a CTA image, and the generated CTA image is good in terms of blood vessel structure reconstruction accuracy, image fidelity and clinical readability.

The invention provides a visual positioning and control method and system for motion of a big arm of a ship loader, and relates to the field of intelligent control. The method comprises the following steps: acquiring multi-modal data; constructing calibration parameters under the global coordinate system based on the multi-modal data, and preprocessing the calibration parameters; taking the preprocessed calibration parameters as input, and outputting multi-modal fusion features and a plurality of assumed poses through a diffusion network; performing virtual rendering on the plurality of hypothetical poses, and calculating a matching score corresponding to each hypothetical pose based on the multi-modal fusion feature and a virtual rendering result; hypothetical poses smaller than a preset screening score in the matching score are obtained to serve as first target hypothetical poses; based on the first target hypothetical pose, calculating a second target hypothetical pose through pose fusion operation; and performing motion control operation on the ship loader big arm based on the second target assumed pose. The problem that a control method in the prior art is low in stability of controlling the pose precision of the large arm of the ship loader is solved.

The invention discloses a tunnel broken surrounding rock grouting reinforcement effect evaluation method. The method comprises the steps that a simulation model is obtained; applying corresponding ground stress; determining the position of a grouting hole; grouting holes are generated; grouting is started; newly-added fractures communicated with the slurry migration channel are merged into the existing slurry migration and diffusion network to update the flow net topological structure; calculating the opening degree of each fracture in the flow net, the flow of slurry and the pressure of the slurry so as to update the slurry pressure and the flow distribution state of the whole slurry migration and diffusion network; the slurry migration and diffusion process is simulated through continuous iterative updating until grouting is finished; updating simulation parameters of the cohesive force unit; the integrity improving effect of tunnel surrounding rock cracks after grouting reinforcement is evaluated; and evaluating the strength improvement effect of the tunnel after grouting reinforcement. According to the method, prediction of the slurry migration and diffusion range, the secondary damage path of secondary excavation disturbance after grouting reinforcement and the bearing capacity is achieved, and the engineering prediction precision and the engineering construction scheme making efficiency are remarkably improved.

The invention relates to the technical field of artificial intelligence and computer graphics, in particular to an image-text semantic proofreading and generative typesetting method and system based on an improved diffusion model, and the method comprises the steps: firstly extracting image-text features through a multi-granularity semantic alignment encoder, and constructing a topological relation; then, a double-flow semantic-layout coupling diffusion network is used for joint generation, and in the reverse denoising process, semantics and layouts are dynamically aligned through cross attention and an energy function; innovatively, a reverse reconstruction proofreading mechanism is introduced in the middle stage of denoising, semantic conflicts and text errors are detected by calculating KL divergence, a large language model is called for correction, and finally, layout is adjusted in a self-adaptive mode, and an optimized document is output. According to the method and the device, generation and proofreading are realized, and the problems of content and form separation, poor semantic consistency and lack of automatic proofreading capability in the prior art are solved.