Method and device for generating three-dimensional grid model by using multiple color pictures

Abstract

The invention provides a method for generating a three-dimensional grid model by using multiple color pictures, which is used for processing a small number of color pictures of different visual angles and corresponding camera parameters based on deformation inference so as to generate a corresponding three-dimensional grid model, and is characterized by comprising the following steps: S1, preprocessing a pre-acquired three-dimensional model data set to obtain a training sample; s2, generating an initial grid template; s3, constructing an image feature extraction network for extracting geometric features and semantic features of the two-dimensional image; s4, constructing a graph convolutional neural network; s5, constructing a loss function; s6, training a three-dimensional model generation model composed of the image feature extraction network and the image convolutional neural network based on the loss function; and S7, inputting the plurality of color pictures and the corresponding camera parameters into the three-dimensional model to generate a model so as to generate the three-dimensional grid model.

Description

technical field [0001] The invention belongs to the field of three-dimensional computer vision, and in particular relates to a method and device for generating a three-dimensional grid model from multiple pictures based on deformation inference. Background technique [0002] 3D shape generation is an important issue connecting images and real space. Thanks to the modeling capabilities of deep neural networks, it has become possible to generate 3D models from only a few pictures. However, generating 3D models only from a single image tends to learn rough shapes, and is not effective for occluded regions and cross-semantic category shapes. In addition, although existing traditional methods can use cross-view information to infer 3D shapes, they require a large number of images with overlapping parts and cannot handle a small number of limited number of images as input. [0003] In recent years, related research work has tried to solve such problems, but due to the diversity o...

AI Technical Summary

Problems solved by technology

[0006] The article [3] uses a triangular mesh (Mesh) to represent a 3D model, which improves the speed and reduces the calculation consumption of the model, but only uses a single image for input, and lacks the correct shape modeling of the occluded area

Images