8results about How to "Guaranteed realism" patented technology

The invention discloses a digital human video generation method and device, electronic equipment and a storage medium. The method comprises the following steps: determining a lip shape and a body language corresponding to a target audio, and generating an initial digital human video according to the lip shape, the body language and a predetermined target digital human; wherein the target digital figure is a digital figure determined according to a user demand; performing super-resolution processing on the initial digital human video through a generative facial prior generative adversarial network model to obtain a target video frame sequence; and generating a target digital human video according to the target video frame sequence and the target audio, and outputting the target digital human video according to a preset format. According to the technical scheme provided by the embodiment of the invention, the generated target digital human video has higher resolution and better video quality, and meanwhile, the facial details of the digital human can be recovered to keep the sense of reality of the image.

The present invention provides a large-scale three-dimensional forest rendering method, including: S1: reading ground elevation data of a three-dimensional forest rendering area and constructing a digital ground model; S2: separating a forest area from the digital ground model and rendering the forest area using a grass noise texture; S3: setting a forest density random distribution function and generating a tree distribution point information map in the forest area; S4: setting tree basic pattern information for each tree distribution point; S5: obtaining tree comprehensive pattern information based on the ground elevation data, tree distribution point information and tree basic pattern information; S6: realizing tree texture display at different viewing angles at close range based on the tree comprehensive pattern information; S7: realizing the display of forest visibility as the observation distance changes using a multi-level of detail technical method. The method can quickly render a large-scale three-dimensional forest, and the rendered three-dimensional forest has a strong sense of reality.

This invention belongs to the technical field of 3D modeling for engineering construction. It discloses a method and system for modeling bulldozer operation maps in mining areas based on 5D Gaussian geometry. The method includes: S1 acquiring RGB images and depth data of the earthmoving operation area using an RGB-D camera; applying a contrast enhancement algorithm to the RGB data and vibration compensation to the depth data; and performing semantic recognition of the category of each pixel location; S2 back-projecting the depth data to generate a 3D point cloud; constructing 3D Gaussian units at the sampling points of the point cloud; and converting them into 5D Gaussian implicit units covering spatial, temporal, and semantic dimensions after fusing semantic category and timestamp information; generating a 3D implicit base map; S3 filtering the 5D Gaussian implicit units; and projecting to obtain a 2D rendered image. This invention provides 3D visualization support for bulldozer operation in mining areas, is applicable to scenarios such as mining, and solves modeling problems such as weak texture in earthmoving and equipment vibration interference.

The present invention provides an image fusion method, device and electronic system. The method includes: registering and aligning multiple images of the same scene to obtain a spatially aligned reference brightness image and at least one underexposed image; wherein the exposure value EV corresponding to the reference brightness image is different from the EV of the underexposed image; determining the effective pixel points in the highlight area based on the color channel parameters corresponding to each pixel point in the highlight area of ​​the underexposed image and the corresponding relationship; determining the calibration parameters of the highlight area based on the color channel parameters corresponding to the effective pixel points in the highlight area, and applying the calibration parameters of the highlight area to calibrate the highlight area to obtain an underexposed optimized image; fusing the reference brightness image and the underexposed optimized image to obtain a fused image corresponding to the same scene, thereby alleviating the problem of color cast in the highlight area of ​​the fused image in the existing method, ensuring the realistic effect of the fused image, and improving the practical value of the fused image.

The invention relates to the technical field of image processing, in particular to a BIM (Building Information Modeling) grid mapping method, which comprises the following steps of: sampling the surface of a BIM, and mapping each point p on the surface into a two-dimensional position UV through a function f (p); mapping a three-dimensional space B containing a definition domain of the function f (p) into a physical space P, then carrying out region division, and independently defining mapping in each region so as to realize that the function f (p) is discontinuous among the regions; skipping blank voxels between the areas, and reducing processing or calculation on the areas; and finally performing mapping application. In the fragment processing process, the evaluation of f (p) only needs to apply scale translation q = sigma (p) and perform single trilinear interpolation texture access at q, and then return to tau (q). This operation replaces linear interpolation of vertex-by-vertex UV in conventional modes, and utilizes trilinear interpolation of texture values supported by native hardware, which is highly optimized in terms of speed and GPU / texture RAM bandwidth.

The invention discloses a self-supervised fine tuning method of a relighting network, and a relighting method and system, and belongs to the technical field of relighting. Starting from an image sample N in the real world, obtaining an illumination condition E of the image sample N; inputting the N and a random illumination condition E * into a pre-trained relighting network to obtain a relighting image N * of the N under the E *; and then taking the relighting image N * as the image input of the relighting network during fine adjustment, E as the illumination condition input of the relighting network during fine adjustment, and N as the training true value of the relighting network, constructing adversarial illumination loss, and performing fine adjustment on the relighting network. According to the method, the paired training data pairs are created from the real world image, it can be ensured that the training truth value of the fine-tuning relighting network is still the real world image, the difference between the fine-tuning relighting network and the real world image is reduced, the generalization of the relighting network is high, and the reality sense of the relighting image output by the network is improved.

The invention discloses an image super-resolution reconstruction method based on a SAUGAN network, and relates to the field of image and machine vision technology of artificial intelligence. The invention integrates a GAN network, a self-attention mechanism, an RRDB+ module, and a U‑Net architecture. A multi-scale feature extraction module is designed in the shallow feature extraction part of the generator, which can retain more fine details of the original image; an RRDB+ module is designed in the deep feature extraction part of the generator, and combined with a self-attention mechanism module, the characteristics of the residual network are fully utilized, the global information of the image is paid attention to, and the global feature information is better utilized for super-resolution image reconstruction, so as to obtain an image with better perceived quality; the discriminator adopts a U‑Net architecture to provide global and local feedback to the generator, improve the local details of the image, and maintain the global and local sense of reality. And spectral normalization regularization is used to stabilize the training process, and finally a good effect of local enhancement and suppression of artifacts is achieved.

The present invention relates to a non-contact operation device for a vehicle and a vehicle, and aims to improve the operation device for the vehicle. The non-contact operation device (40) of the vehicle (1) comprises: a projection device (43) that projects an image onto the vehicle interior (3) of the vehicle (1); a generation unit (51) that generates and updates the image projected by the projection device (43); a stimulus response output unit (53) that outputs tactile stimulation caused by an ultrasonic field as a response to the operation of the operating part of the occupant who operates the image projected on the vehicle interior (3); and a detection device (44) that detects the operating part of the occupant who performs non-contact operation on the image projected on the space of the vehicle interior (3). The generation unit (51) and the stimulus response output unit (53) update the output of the image and the tactile stimulation based on the operation information of the operating part of the occupant detected by the detection device (44).