Three-dimensional reconstruction method based on single image

Abstract

The invention discloses a three-dimensional reconstruction method based on a single image. The method comprises a training stage, wherein a training sample comprises sampling points on a three-dimensional model, actual implicit function values corresponding to the sampling points and the single image corresponding to the three-dimensional model; a feature space is generated through a single input picture, sampling points of the three-dimensional model are used as query points, implicit function values of the query points are predicted, loss is calculated according to the predicted implicit function values and actual implicit function values, and fitting of an implicit function field is completed; in the reconstruction stage, after the training stage is completed, a feature space is generated for a single input picture, a plurality of points are sampled in an observation space, and the points form query points; predicting an implicit function value of a query point; and inputting the query point and the prediction implicit function value into a marching cubes algorithm, and reconstructing to obtain a three-dimensional model. According to the method, the problem of serious lack of prior conditions in single image reconstruction is solved, and fine details can be obtained in a visible part.

Description

technical field [0001] The invention belongs to the field of computer vision and image processing, and relates to a method for three-dimensional reconstruction of a two-dimensional image, in particular to a deep learning reconstruction method based on a single picture. Background technique [0002] Due to blurriness (e.g. the presence of parallax), single vision Figure three Dimension reconstruction is an ill-posed problem. Unlike 3D reconstruction based on multi-view images, single-view Figure three Dimensional reconstruction requires strong prior knowledge. The current mainstream single Figure three Dimensional reconstruction methods are mainly divided into template-based explicit methods and non-template-based implicit methods. The implicit methods can be further divided into voxel-based methods, point cloud-based methods, and implicit function-based methods. Each method has its own advantages and disadvantages. The current research hotspot is the implicit function m...

AI Technical Summary

Problems solved by technology

The current research hotspot is the implicit function method. There are several common problems in this type of method: (1) The generated reconstruction results lack details; (2) The missing of important parts caused by the lack of a priori

Images