CN-122019825-A - Digital material selection method and material selection equipment for AIGC home design scene

CN122019825ACN 122019825 ACN122019825 ACN 122019825ACN-122019825-A

Abstract

A AIGC home design scene-oriented digital material selection method and material selection equipment relate to the technical field of artificial intelligence and are used for improving the accuracy of selecting physical entity materials according to AIGC images. In the method, the material selecting device obtains the intrinsic texture topological tensor representing the texture structure by identifying and quantizing AIGC non-physical rendering style fingerprints of the image and removing the non-physical rendering style fingerprints from the feature layer, and then aligns the intrinsic texture topological tensor to the feature space of the real physical building material through the preset domain generalization feature mapping network, so that the interference of AIGC rendering style is ignored in the forced retrieval process, the comparison of the texture topological structure is focused, and the accuracy of selecting the physical entity material according to the AIGC image is improved.

Inventors

HUANG YANHONG
PAN JIN
Fang Luhuan
HUANG ZHENGMIN
WANG QIN
YANG CHENGHAO

Assignees

暗壳科技(深圳)有限公司

Dates

Publication Date: 20260512
Application Date: 20260206

Claims (10)

1. The digital material selecting method for AIGC home design scenes is characterized by being applied to material selecting equipment and comprises the following steps: Performing texture analysis on the AIGC virtual rendering image to be retrieved to obtain a target texture region in the AIGC virtual rendering image; performing multi-scale convolution operation on the target texture region according to a preset cascade convolution filter group to obtain a high-dimensional characteristic response tensor comprising a plurality of channel layers; Respectively calculating a pixel response mean value and a pixel response standard deviation of each channel layer along the channel dimension of the high-dimensional characteristic response tensor; determining a set consisting of the pixel response mean values and the pixel response standard deviations of all channel layers as a non-physical rendering style fingerprint, wherein the non-physical rendering style fingerprint characterizes non-physical illumination dispersion distribution and artistic filter tone distribution in the AIGC virtual rendering image; in each channel layer of the high-dimensional characteristic response tensor, subtracting the pixel response mean value corresponding to the channel layer from the pixel response value of the channel layer, and dividing the pixel response mean value by the pixel response standard deviation corresponding to the channel layer to obtain an intrinsic texture topology tensor from which the influence of the non-physical rendering style fingerprint is removed; Inputting the intrinsic texture topology tensor into a preset domain generalized feature mapping network to perform feature transformation to obtain a texture feature vector to be retrieved; Calculating cosine similarity values between the texture feature vectors to be searched and preset standard texture fingerprint vectors of all stock material units in a preset physical building material database; and screening out physical entity material information matched with the target texture region in texture topological structure from the physical building material database according to the sequence of the cosine similarity values from large to small.
2. The method according to claim 1, wherein the constructing of the preset domain generalization feature mapping network specifically comprises: obtaining a real physical building material sample and AIGC virtual rendering samples from a sample library; determining an intrinsic texture topology tensor of the real physical building material sample and a non-physical rendering style fingerprint of the AIGC virtual rendering sample; Performing inverse standardization operation on the intrinsic texture topology tensor of the real physical building material sample based on the non-physical rendering style fingerprint of the AIGC virtual rendering sample to obtain a synthesized feature tensor with AIGC style noise; inputting the synthesized feature tensor into a feature mapping network to be trained to obtain a prediction result; calculating a classification loss function and a measurement learning loss function based on the prediction result and the texture class label of the real physical building material sample; And reversely updating network parameters of the feature mapping network based on the classification loss function and the measurement learning loss function until the network converges to obtain the generalization feature mapping network of the preset domain.
3. The method according to claim 2, wherein the step of performing an inverse normalization operation on the intrinsic texture topology tensor of the real physical building material sample based on the non-physical rendering style fingerprint of the AIGC virtual rendering sample to obtain a synthesized feature tensor with AIGC style noise specifically comprises: Determining a first texture semantic category of the real physical building material sample, and selecting AIGC virtual rendering samples with a second texture semantic category from the sample library, wherein the first texture semantic category is different from the second texture semantic category; And carrying out affine transformation on the intrinsic texture topology tensor of the real physical building material sample based on the non-physical rendering style fingerprint of the AIGC virtual rendering sample with the second texture semantic category to obtain a synthesized feature tensor with cross-domain style conflict.
4. The method according to claim 1, wherein the construction of the preset standard texture fingerprint vector for each stock material unit in the preset physical building material database specifically comprises: Acquiring physical photographed images of all stock material units in the preset physical building material database; determining an intrinsic texture topology tensor for each of the physical live images; inputting the intrinsic texture topology tensor corresponding to each stock material unit into the preset domain generalization feature mapping network to perform feature transformation to obtain a preset standard texture fingerprint vector of each stock material unit.
5. The method according to claim 1, wherein the step of performing texture analysis on the AIGC virtual rendered image to be retrieved to obtain the target texture region in the AIGC virtual rendered image specifically includes: identifying a home plane structure in the AIGC virtual rendered image; Extracting an edge line set based on the home plane structure; calculating perspective vanishing points and homography matrixes of the AIGC virtual rendering images according to the edge line sets; Performing inverse perspective projection transformation on the AIGC virtual rendering image based on the homography matrix to obtain a corrected orthographic projection image; and cutting out an area meeting a preset resolution threshold from the orthographic projection image to serve as the target texture area.
6. The method according to claim 1, wherein the step of inputting the intrinsic texture topology tensor into a predetermined domain generalization feature mapping network to perform feature transformation to obtain a texture feature vector to be retrieved specifically comprises: the preset domain generalization feature mapping network comprises a channel attention module and a multi-layer perceptron module which are sequentially connected; Performing global average pooling on the intrinsic texture topology tensors to obtain channel descriptor vectors; inputting the channel descriptor vector into the channel attention module to obtain a weight coefficient corresponding to each channel; weighting and reorganizing each channel layer of the intrinsic texture topology tensor based on the weight coefficient to obtain a weighted feature vector; and inputting the weighted feature vector into the multi-layer perceptron module for nonlinear mapping to obtain the texture feature vector to be retrieved.
7. The method of claim 1, wherein after the step of screening the physical building material database for physical entity material information matching the target texture region in texture topology in order of the cosine similarity value from high to low, the method further comprises: Calculating a first local color histogram of the target texture region in an HSV color space; calculating a second local color histogram of the physical photographed image of each piece of the screened physical entity material information; calculating a barking distance between the first local color histogram and the second local color histogram; And reordering the screened physical entity material information according to the sequence from the small value to the large value of the Babbitt distance to obtain a final physical entity material list.
8. A material selection device comprising one or more processors and memory coupled to the one or more processors, the memory for storing computer program code comprising computer instructions that the one or more processors invoke to cause the material selection device to perform the method of any of claims 1-7.
9. A computer program product containing instructions, which, when run on a material selection device, cause the material selection device to perform the method of any one of claims 1-7.
10. A computer readable storage medium comprising instructions which, when run on a material selection device, cause the material selection device to perform the method of any one of claims 1-7.

Description

Digital material selection method and material selection equipment for AIGC home design scene Technical Field The application relates to the technical field of artificial intelligence, in particular to a digital material selecting method and material selecting equipment for AIGC home design scenes. Background With the wide application of the technology of generating Artificial Intelligence (AIGC) in the fields of home design and digital interior decoration, designers can quickly generate virtual decoration effect figures containing rich texture details by utilizing the artificial intelligence. Based on AIGC images, physical entity materials are retrieved and matched, and the related technology mainly adopts an image feature matching technology based on deep learning. Specifically, the technique first deploys a convolutional neural network that is primarily pre-trained based on a real photographic image dataset as a feature extraction network. In the retrieval process, the image to be retrieved is input into the feature extraction network, and the feature extraction network maps the image into a high-dimensional feature vector according to color distribution, edge gradient and texture details under real optical imaging. And then, calculating the numerical value between the feature vector of the query image and the feature vector of each commodity image pre-stored in the commodity library by using a cosine similarity algorithm. And performing descending order arrangement according to the calculated similarity value, and recommending physical materials corresponding to the commodity image with the highest similarity to a user as a matching result. However, when a customer sees a certain wood grain texture in AIGC images, it is necessary to retrieve and match physical and physical materials similar in visual characteristics to the virtual texture in the effect map from the real building material commodity database, since AIGC images are composite images generated essentially based on probability distribution, they contain non-physical rendering characteristics (such as non-physical light diffusion distribution, etc.). Because the feature extraction network of the related art is based on the training weight of the real photographic image, the convolution kernel can misidentify the specific rendering characteristics of the AIGC images as the effective content features of the images by capturing the pixel statistics rule, and encode the style information of the non-textures into the output feature vectors, so that the retrieval result tends to recommend physical entity materials with similar hue or blurring degree and unmatched texture topological structures (such as wood grain trend), and the accuracy of selecting the physical entity materials according to the AIGC images is reduced. Disclosure of Invention The application provides a AIGC household design scene oriented digital material selection method and material selection equipment, which are used for improving the accuracy of selecting physical entity materials according to AIGC images. A digital material selecting method for AIGC home design scenes is characterized by being applied to material selecting equipment, and comprises the steps of carrying out texture analysis on AIGC virtual rendering images to be retrieved to obtain target texture areas in AIGC virtual rendering images, carrying out multi-scale convolution operation on the target texture areas according to a preset cascade convolution filter set to obtain high-dimensional feature response tensors comprising a plurality of channel layers, respectively calculating pixel response mean values and pixel response standard deviations of each channel layer along channel dimensions of the high-dimensional feature response tensors, determining a set consisting of the pixel response mean values and the pixel response standard deviations of all the channel layers as non-physical rendering style fingerprints, wherein the non-physical rendering style fingerprints represent non-physical illumination dispersion distribution and artistic filter tone distribution in AIGC virtual rendering images, subtracting the pixel response mean values of the channel layers from the pixel response values of the channel layers in each channel layer of the high-dimensional feature response quantities, then dividing the pixel response standard deviations of the channel layers to obtain topological feature texture vectors which are influenced by the non-physical feature vectors, mapping the feature vectors to be compared with the pre-set feature vectors in a pre-set network, and obtaining the texture feature vectors which are similar to the physical texture vectors of the texture vectors to be retrieved from a pre-set network storage object texture feature map data map to the map values of the pre-set feature map data to the physical feature map data map values to the feature map data to the physical feature map data to the map fe