139 results about "Drop shadow" patented technology

The invention discloses a three-dimensional scene target blanking method and system based on 3DGS, and belongs to the field of computer vision processing. The method comprises the following steps: firstly, performing three-dimensional reconstruction through a multi-view image to generate an original Gaussian point cloud scene; after a scene is divided into foreground and background point clouds based on target text information, spatial expansion is performed on the foreground to generate a three-dimensional soft mask. And projecting the mask to a two-dimensional image space to generate a two-dimensional mask image of each view angle, and inputting the two-dimensional mask image and an original image into an image blanking model to generate a candidate blanking image set. After a user selects a reference image from an initial view angle, a candidate image most similar to a preorder view angle is selected as a final blanking result from a non-initial view angle, and fidelity enhancement is carried out. And finally, optimizing the original Gaussian point cloud parameters by using the blanking result of each view angle to obtain a target three-dimensional scene after blanking. Based on the invention, a user can carry out texture remodeling on the structure of the original three-dimensional scene, and the multi-view consistency of the three-dimensional scene is maintained.

The invention is suitable for the technical field of sentiment analysis, and particularly relates to a multi-modal sentiment analysis method based on pre-training and text modal guidance, and the method is implemented in two stages: the pre-training stage comprises the steps of obtaining initial features of each modal of multi-modal data, and decomposing each modal feature into similar features and different features; constructing positive and negative pairs in and between samples, calculating comparison loss and single-modal prediction errors, extracting module and projector parameters, and inputting a short video to obtain each modal feature; high-scale text representation is learned through initial text similar features, super-modal features are adaptively updated through a multi-head attention mechanism, and audio and video redundant information is suppressed by using a weight module guided by a text mode; through the data sampler and the feature comparison learning method, multi-mode fusion with the text mode as an anchor point is achieved, the emotional tendency in a complex scene can be recognized more accurately, and the method has wide value in the fields of emotion classification, social media content recommendation and the like.

The invention discloses a dynamic scene robust vision SLAM (Simultaneous Localization and Mapping) method based on multi-feature collaborative optimization, which comprises the following steps of: acquiring an image sequence, carrying out dynamic target detection and segmentation through an instance segmentation network, generating a segmentation mask containing a dynamic region mark, and identifying and separating a dynamic object and a static background; removing feature points corresponding to the dynamic object based on the segmentation mask to obtain static feature points; carrying out pose estimation based on the static feature points, and for the key frame, carrying out feature matching with the previous key frame by minimizing a re-projection error, and solving to obtain the camera pose of each key frame; for non-key frames, performing camera pose tracking and data association on the previous frame by adopting an optical flow algorithm, and accumulating solving results to obtain pose tracks of all the non-key frames; the key frames and the non-key frames are subjected to differential processing by fusing feature matching and an optical flow algorithm, so that the calculation efficiency is remarkably improved while the positioning precision is ensured, and the real-time performance is improved.

The invention discloses a compact multi-modal large model realized through embedded visual compression, and aims to reduce the calculation overhead of the model under high-resolution input and maintain full expression of image details through an embedded visual compression strategy. The compact multi-modal large model comprises the following steps: extracting visual features through a visual encoder; text features are obtained through a word segmentation device; the visual features are processed into compact visual features by a space attention pooling module; performing linear projection on the compact visual features and the text features by a projector, inputting the compact visual features and the text features into a multi-modal large model, and filling # imgabs0 # learnable query features at the tail of a sequence; the compact visual features, the text features and the query features participate in multi-modal interaction before a # imgabs 1 # layer of the multi-modal large model, and image information related to an instruction is captured through an embedded query module at a # imgabs 2 # layer so as to supplement image details lost in the attention pooling process.

The invention provides a multi-modal large model method based on layered visual injection and a mixed attention mechanism. The method comprises the following steps: firstly, extracting a visual feature sequence of an input image through a pre-trained visual encoder, and projecting the visual feature sequence to a dimension which is the same as that of a large language model embedding space; and then, in each layer, performing efficient fusion on the visual features and the text features by using a mixed attention mechanism, and generating a comprehensive key value sequence, thereby realizing cooperative processing of visual and text information. Compared with an existing multi-mode large model, the method effectively solves the problem that the calculation complexity is remarkably improved due to the fact that the sequence length is increased. Through layered visual injection and a mixed attention mechanism, the overall calculation cost is remarkably reduced while the model performance is maintained, and the multi-modal large model calculation efficiency is improved.

The invention discloses a text-to-three-dimensional Gaussian shape generation method and device, equipment and a medium, and the method comprises the steps: obtaining an initialized three-dimensional Gaussian model through a pre-trained three-dimensional generator according to the text input for generating a target task; performing multi-view projection to obtain a key view image and a group of auxiliary view images; performing feature extraction on the auxiliary view image, and fusing the extracted features with the features of the key view image through an attention mechanism; performing multi-view joint geometric optimization on the initialized three-dimensional Gaussian model through fractional distillation loss by using the fused features and text input injection diffusion model to obtain a three-dimensional Gaussian model subjected to geometric optimization; and a layered Gaussian constraint mechanism is implemented, reinforced constraint is applied to Gaussian points generated earlier, higher degree of freedom is given to Gaussian points newly added in the later period, and a final three-dimensional Gaussian shape is obtained. According to the method, the geometric semantic consistency and the overall fidelity and fineness of the generated result under different visual angles are improved.

The invention discloses a text-guided 3D knitted product model generation and editing method and device, and relates to the field of computer vision, and the method comprises the steps: S1, employing a 3D diffusion model to enable a knitted product prompt text to generate an initial point cloud; s2, optimizing the initial point cloud and initializing the initial point cloud into a 3D Gaussian; s3, performing projection rendering on the initial 3D Gaussian to obtain a multi-view image sequence, and performing semantic segmentation on the multi-view image sequence based on an editing prompt text to obtain a mask sequence; s4, acquiring a 3D Gaussian with a semantic tag based on the mask sequence; s5, projecting the 3D Gaussian with the semantic tag to a 2D plane to obtain a rendered image; inputting the rendered image and the editing prompt text into a 2D diffusion model, and outputting a loss gradient; and utilizing the loss gradient to guide the 3D Gaussian to carry out iteration, wherein the 3D Gaussian obtained after iteration is finished is the final 3D knitted product model. According to the method, the 3D Gaussian is introduced, the semantic tag is added for the 3D Gaussian by using the editing prompt text, and the generation efficiency, the local semantic control and the high-precision editing are considered.

The invention discloses a graph correction method, a storage medium and a storage terminal.The graph correction method comprises the steps that a to-be-corrected layout is provided, and the to-be-corrected layout comprises a plurality of initial to-be-corrected graphs; a first initial to-be-corrected graph and a second initial to-be-corrected graph are obtained from the multiple initial to-be-corrected graphs, and the projection of the first initial to-be-corrected graph and the projection of the second initial to-be-corrected graph in the first direction and the projection of the second initial to-be-corrected graph in the second direction do not coincide; the first direction is perpendicular to the second direction, the first direction and the second direction are parallel to the surface of the to-be-corrected layout, the first initial to-be-corrected graph is provided with a first vertex, the second initial to-be-corrected graph is provided with a second vertex, and the first vertex is opposite to the second vertex; part of the initial to-be-corrected graph is removed, to-be-corrected graphs are generated, the removed area of the initial to-be-corrected graph comprises one or two of a first vertex and a second vertex, and the distance between every two adjacent to-be-corrected graphs meets the requirement of a mask limiting rule; and performing optical proximity effect correction on the to-be-corrected pattern. According to the graph correction method, the precision of the graph is improved.

The invention discloses a real-time dynamic light and shadow rendering optimization method based on an unreal engine, and relates to the field of light and shadow rendering, and the method comprises the steps: firstly distributing shadow importance scores for all grid objects in a scene based on a camera view cone and AI context data, and carrying out the priority ranking of potential shadow projectors; and then, performance budget evaluation is carried out in combination with the current frame rate and the GPU load, and intelligent shadow projector selection and LOD application are carried out on the sorted shadow projection object list according to the performance budget evaluation. And finally, the optimized shadows to be rendered are sent to a light visual angle rendering and VSM page filling process in batches, and a movable light source optimization virtual schlieren is generated. According to the method, the unnecessary calculation overhead is remarkably reduced, so that the rendering efficiency and the frame rate stability are greatly improved while the high-quality dynamic light and shadow effect is ensured, and the core problem that the traditional performance and quality are difficult to balance is effectively solved.

The invention relates to the technical field of virtual reality rendering, in particular to a virtual scene rendering method and device, terminal equipment and a medium. The method comprises the following steps: loading a pre-constructed preform assembly; the preform assembly comprises at least one dynamic object and at least one static object in the scene to be rendered; illumination information of all static objects is bound on the prefabricated body assembly; performing shadow rendering on the at least one static object according to the bound illumination information; and performing shadow rendering on the at least one dynamic object by adopting a plane shadow projection algorithm. According to the method, the illumination baking information is directly mounted on the preform assembly, so that the illumination information can be obtained without baking the static object in the scene again, shadow rendering is performed on the dynamic object by adopting the plane shadow projection algorithm, the dynamic object also has real-time illumination, and the scene is more real. Therefore, the dynamic shadow and the baking shadow in the existing virtual scene can be effectively fused, and the playing experience of a low-configuration equipment user is improved.

The invention provides a path fitting method, device and equipment based on a digital three-dimensional scene, and the method comprises the steps: receiving personalized parameters defined by a user at a front end, and packaging the personalized parameters into a data volume which can be analyzed by a three-dimensional graphic rendering engine; based on a three-dimensional graphic rendering engine, analyzing the data volume and converting the data volume into parameters required for creating a target path; generating a mapping component of a ground projection effect between the corresponding data point positions based on the converted parameters; the starting point position of the target path is calculated, and the position, rotation and scaling of the mapping assembly are determined; and based on the length and direction of the target path, calculating the mapping of each section of path forming the target path in the whole path texture mapping space so as to realize the visual effect of displaying the texture style and moving the covering along the target path. According to the scheme, the problem that the path is easy to penetrate or suspend can be solved, meanwhile, the accuracy and the visual effect of path creation are improved, and the flexibility and the response speed of the system are enhanced.

The invention relates to the technical field of virtual reality, and discloses a live-action three-dimensional scene-based video fusion method and electronic equipment, and the method comprises the following steps: obtaining the three-dimensional data of a live-action scene through a data collection device, and generating a three-dimensional scene model through a point cloud reconstruction or multi-view geometric method; optimizing the three-dimensional scene model, including grid simplification, texture mapping and light source information initialization, and obtaining the spatial position and radiation intensity distribution of a scene light source. The Passon equation is combined with the variational optimization technology, and precise modeling of illumination distribution of the complex three-dimensional scene is achieved; high-precision modeling and motion trail extraction of a dynamic target are completed through a deep learning segmentation and depth estimation technology; based on Lambert reflection and projection correction, the consistency of a dynamic target light shadow and a scene light field is ensured; in combination with multi-resolution light and shadow optimization and GPU acceleration, the real-time rendering performance is remarkably improved, and the problems of abrupt light and shadow, untrueness and insufficient real-time performance in the prior art are solved.

The invention provides a visual SLAMMOT system and method fusing object plane features, and introduces an object plane feature extraction and matching method based on a normal vector on the basis of traditional point features. A plane region is judged by analyzing the consistency of normal vectors of adjacent points, and stable and accurate extraction and association of multi-plane features in a multi-frame and multi-object scene are realized by combining a point feature matching result and plane parameter similarity. In addition, the invention further provides a pose optimization method fusing object plane constraints, object map point re-projection, an object constant-speed motion model, point-to-plane constraints and plane-to-plane constraints are incorporated into a factor graph optimization framework, and the precision of object surface reconstruction and pose estimation is further improved. According to the method, by fusing object plane features, the performance of a visual SLAMMOT system in a dynamic environment is remarkably enhanced, and the method has high practicability and innovativeness.

The invention discloses an image processing method and device, a storage medium, equipment and a program product, and the method comprises the steps: obtaining a cue word and a scene grid, the cue word being text information which is provided by a user and represents a scene style, and the scene grid being a three-dimensional grid which is generated based on real scene reconstruction and has real textures; generating a stylized panoramic texture map at the central position of the scene grid according to the prompt word; projecting the texture of the stylized panoramic texture map to a visible area of the scene grid to obtain a first stylized grid texture map; and performing spatial texture propagation processing on the first stylized grid texture map to fill an invisible area of the scene grid to obtain a second stylized grid texture map. According to the method, the stylized grid texture map can be automatically generated according to the prompt word provided by the user, meanwhile, the space consistency and the map correctness are ensured, and a complex geometric structure in a real scene can be completely covered.

The invention discloses an immersive design experience method and system based on VR, and relates to the technical field of virtual reality, and the method comprises the steps: monitoring the position and attitude data of a dynamic object in a scene in real time between continuous rendering frames of the virtual reality scene, the position and attitude data comprising a three-dimensional coordinate, a rotation angle and a scaling; and when it is monitored that the position posture data of the dynamic object changes, determining a first projection area corresponding to the dynamic object in a previous frame rendering cache in a scene based on the position posture data before the change, and defining the first projection area as a shadow erasing area. According to the method, by dynamically monitoring the position and posture change of the object, accurately delimiting the shadow erasing area and the reconstruction area, and only executing shadow clearing and generation calculation on the local area, invalid operation is greatly reduced, calculation complexity is remarkably reduced, rendering efficiency is improved, and shadow updating delay and picture lagging are avoided.

The application discloses a video shadow detection method and device based on a dynamic prompt memory network, which can process confused shadows with the help of a projected shadow object, can aggregate time information without accumulating errors, and comprises a dynamic prompt module and a prompt-based memory module; the dynamic prompt module converts semantic information provided by a visual base model DINOv2 into a shadow mask when direct encoding of a shadow is difficult, local prompting focuses on semantic conversion and identifies a shadow position based on semantic information, and global prompting identifies a shadow boundary based on color and texture information; the memory module is used for solving deformation and long-term time consistency problems, preventing error accumulation, using local prompting as a time matching agent, reducing memory usage, and minimizing dependence on historical shadow masks; and the method can improve detection performance in a complex scene and maintain time consistency of results during long-term detection.

The present disclosure provides a shadow rendering method, apparatus, computer equipment and storage medium for a three-dimensional model, wherein the method comprises: determining a shadow body corresponding to a three-dimensional model in a virtual three-dimensional space; the shadow body is used to indicate a projection rendering range of the three-dimensional model when projecting the three-dimensional model onto a projection surface of a preset object in the virtual three-dimensional space; in response to at least one target vertex on the projection surface being located within the area where the shadow body is located, determining a display shadow intensity corresponding to the target vertex based on second position information of the target vertex in the virtual three-dimensional space and first position information corresponding to the shadow body; rendering the projection surface based on the display shadow intensity corresponding to the at least one target vertex to obtain a shadow rendering result of the three-dimensional model on the projection surface.

The application discloses a sketch representation enhancement method and system based on self-supervised learning, in the image and text feature extraction stage, the sketch is augmented and transformed, the text box content in the sketch is recognized, and the text content and the sketch are encoded respectively to obtain text features and image features; in the sketch representation enhancement stage under the guidance of the text, the text features are used as the basis, and a guide attention unit is applied to enhance the image features; in the contrast self-supervised learning stage, the original image and the augmented sketch enhancement features are mapped into a low-dimensional vector space through a projection function, and the loss is calculated by the low-dimensional vector and the model is optimized.

The invention discloses a training-free gridding video feature compression method. The long video processing efficiency of a multi-modal large model is remarkably improved through space-time fusion and semantic compression technologies. Comprising the following steps: recombining a long video frame sequence into a grid image according to a time sequence, and extracting spatio-temporal joint features by using a visual Transform; calculating cosine similarity based on the global semantic center, and screening the first U key visual marks; and non-key mark information is dynamically fused to the key mark through the normalized weight, so that redundant compression is realized. And finally, aligning the compressed visual mark with a text space through linear projection, splicing the compressed visual mark with task prompt and user query, and inputting the spliced result into a large language model. According to the method, on the premise of being completely free of training, the number of visual marks is reduced, a GPU video memory is reduced, the reasoning speed is increased, meanwhile, the accuracy is improved, the problem of calculation bottleneck in long video understanding is effectively solved, and the method is suitable for plug-and-play deployment of mainstream multi-mode frames such as LLaVA and Video-LLaMA.

In view of the problems that when there are dynamic objects in the scene, a visual SLAM system has a large gap between the positioning result and the true value, and the system robustness is low. Based on the traditional visual odometer, a binocular SLAM front-end optimization algorithm that is robust in a dynamic environment is proposed. The algorithm first dedistorts the binocular image, then extracts and triangulates the image, and pre-screens the dynamic feature points according to the visual geometric constraints combined with the reprojection error chi-square distribution model. On this basis, a multidimensional space directed graph model is constructed by using the information of each feature point itself and the motion consistency of the feature points on the same dynamic object, and an AP clustering algorithm is introduced to accurately screen the feature points, thereby accurately filtering out the point groups that do not meet the static assumption. The overall positioning trajectory of the algorithm is closer to the theoretical value, and can effectively improve the operation accuracy and robustness of the SLAM system in dynamic scenes.

1. The name of the design product: projector. 2. The use of the design product: for the projection and play of image, video and the output of audio. 3. The design points of the design product: in shape. 4. The picture or photo that can best show the design points: perspective view 1.

Techniques for improving how LSR is performed are disclosed. A service accesses a depth image, a pose correction matrix, and a color image. The service extracts a carrier geometry from the depth image. The service forward projects the LSR carrier geometry by multiplying each vertex of the LSR carrier geometry with the pose correction matrix. While the GPU is operating in a shadow map mode, the service causes the GPU to perform a rasterization process to produce a UV map and a Z buffer. The service discards the UV map, resulting in the per pixel UV corrections included in the UV map also being discarded. The service recovers the per pixel UV corrections using the Z buffer. The service uses the recovered per pixel UV corrections to resample the color image, resulting in generation of a corrected color image.

1. The name of the present design product: projector (five series two). 2. The use of the present design product: for projection. 3. The design points of the present design product: in shape. 4. The picture or photo that can best show the design points: perspective view.

This invention discloses a method and device for displaying protruding shapes on a background wall in a 2D plane, belonging to the field of home decoration design technology. By projecting grouped data of the protruding area, the modeling surface, and the light trough surface onto a two-dimensional plane, the invention displays the position and outline information of the protruding shapes on the background wall on a 2D plane. This solves the problem of being unable to represent or unclearly representing the protruding shapes on a 2D plane. This allows users to obtain the relative position and other relationships between design elements and the background wall modeling during design, preventing them from interfering with each other, thereby improving the user's design experience and efficiency.

1. The name of the design product: projector (FS210P cloud platform B). 2. The use of the design product: for projection. 3. The design points of the design product: in shape. 4. The picture or photo that best indicates the design points: perspective drawing.

This invention provides a zero-shot anomaly detection method and system based on bidirectional alignment enhancement. The method includes: achieving direct interaction between global and local features through the cross-attention module CAGL; extracting fine-grained local features through the region token alignment module RegionAlign; and projecting global visual semantics onto the text embedding space through the text embedding enhancement module TextAug to generate domain-aware cue words. This invention solves the technical problems of poor generalization ability, neglect of local spatial details, and difficulty in hierarchical modeling of multi-scale visual anomalies.

1. The name of the design product: the front frame part of a projector. 2. The use of the design product: the product as a whole is a projector for projecting video, images, etc.; the part to be protected is the front frame part of the projector. 3. The design points of the design product: the shape shown by solid lines in the perspective view. 4. The picture or photo that best indicates the design points: the perspective view. 5. Other circumstances that need to be explained: the design is a partial design, the part shown by solid lines is the part to be protected as a partial design, and the dashed line is only a line indicating the boundary between the part to be protected as a partial design and other parts. The part other than the part marked in gray in the reference figure is the part to be protected.

1. The name of this design product: Projection information dual-light fusion display graphical user interface for electronic equipment. 2. Purpose of this design product: an electronic device. 3. The key design point of this design product lies in the graphical user interface. 4. The picture or photo that best illustrates the key points of the design: Figure 4 showing the interface change states of Design 1. 5. This design product is a flat product and other views are omitted. 6. Designate Design 1 as the base design. 7. Purpose of the graphical user interface: The graphical user interface of this design is used to display projection information through dual-lamp fusion. Click "Pixel Headlight" in the main view of Design 1 to enter the Design 1 interface change state diagram 1. Click "My Projection" in Design 1 interface change state diagram 1 to enter Design 1 interface change state diagram 2. In the Design 1 interface change state diagram 2, click the "+" on the right side of "My Projection" to enter the Design 1 interface change state diagram 3. Click the "Text Projection" area in Design 1 Interface Change Status Diagram 3 to enter Design 1 Interface Change Status Diagram 4. After editing the text in Design 1 Interface Change State Diagram 4, click the play button below the edited text to enter Design 1 Interface Change State Diagram 5. In the Design 1 interface change state diagram 5, click the setting icon in the upper right corner to enter the Design 1 interface change state diagram 6. Click "One-click fusion" in Design 1 interface change status diagram 6 to enter Design 1 interface change status diagram 7. After the dual lights are successfully integrated, the interface changes from the design 1 state diagram 7 to the design 1 state diagram 8. Among them, the text edited in the design 1 interface change state diagram 4 will be projected externally. Click "Pixel Headlight" in the Design 2 main view to enter the Design 2 interface change state diagram 1. Click "My Projection" in Design 2 interface change state diagram 1 to enter Design 2 interface change state diagram 2. In the Design 2 interface change state diagram 2, click the "+" on the right side of "My Projection" to enter the Design 2 interface change state diagram 3. In the Design 2 interface change state diagram 3, click the "Graffiti Projection" area to enter the Design 2 interface change state diagram 4. After editing the graffiti pattern in Design 2 Interface Change State Figure 4, click the play button below the edited graffiti pattern to enter Design 2 Interface Change State Figure 5. In the Design 2 interface change state diagram 5, click the setting icon in the upper right corner to enter the Design 2 interface change state diagram 6. Click "One-click fusion" in the Design 2 interface change status diagram 6 to enter the Design 2 interface change status diagram 7. After the dual lights are successfully integrated, the interface changes from the design 2 state diagram 7 to the design 2 state diagram 8. Among them, the graffiti pattern edited in the interface change state diagram 4 of Design 2 will be projected externally. The graphical user interface of this design is used to display projection information. Click "Pixel Headlight" in the Design 3 main view to enter the Design 3 interface change state Figure 1. Click "My Projection" in Design 3 interface change state diagram 1 to enter Design 3 interface change state diagram 2. In the Design 3 interface change state diagram 2, click the "+" on the right side of "My Projection" to enter the Design 3 interface change state diagram 3. Click the "Text Projection" area in Design 3 Interface Change Status Diagram 3 to enter Design 3 Interface Change Status Diagram 4. After editing the text in Design 3 Interface Change State Diagram 4, click the play button below the edited text to enter Design 3 Interface Change State Diagram 5. In the Design 3 interface change state diagram 5, click the setting icon in the upper right corner to enter the Design 3 interface change state diagram 6. Click "One-click fusion" in Design 3 interface change status diagram 6 to enter Design 3 interface change status diagram 7. After the dual lights are successfully integrated, the interface changes from the design 3 state diagram 7 to the design 3 state diagram 8. Among them, the text edited in the design 3 interface change state diagram 4 will be projected externally. Click "Pixel Headlight" in the Design 4 main view to enter the Design 4 interface change state Figure 1. Click "My Projection" in Design 4 interface change state diagram 1 to enter Design 4 interface change state diagram 2. In the Design 4 interface change state diagram 2, click the "+" on the right side of "My Projection" to enter the Design 4 interface change state diagram 3. In the Design 4 interface change state diagram 3, click the "Graffiti Projection" area to enter the Design 4 interface change state diagram 4. After editing the graffiti pattern in Design 4 Interface Change State Figure 4, click the play button below the edited graffiti pattern to enter Design 4 Interface Change State Figure 5. In the Design 4 interface change state diagram 5, click the setting icon in the upper right corner to enter the Design 4 interface change state diagram 6. Click "One-click fusion" in the Design 4 interface change state diagram 6 to enter the Design 4 interface change state diagram 7. After the dual lights are successfully integrated, the interface changes from the design 4 state diagram 7 to the design 4 state diagram 8. Among them, the graffiti pattern edited in the design 4 interface change state diagram 4 will be projected externally.