Canonical-Space 3D Model Reconstruction From a Single Image
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model reconstruction methods based on single-view pictures face challenges in efficiency and generalization, with some requiring multiple views and others being limited in reconstruction quality.
Innovation Solution
A method that converts sampling points from a target space to a canonical space using object pose-shape parameters, fusing global and pixel-level features to predict a 3D Gaussian parameter from a single 3D picture, enabling accurate reconstruction with reduced data requirements and improved generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple views are used for model reconstruction, then reconstruction quality is improved, but input data requirements and processing complexity increase
Solution Approach 1:
The patent transforms the input data from multiple view images into a single-view 3D image with enhanced depth information. By changing the parameter representation from 2D multi-view to 3D single-view with depth maps, the system achieves high reconstruction quality while reducing input data requirements to just a single picture.
2Measurement precision
If complex reconstruction algorithms are used, then model accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-processing the single input image to generate a 3D representation with depth information before the actual reconstruction process. This preliminary 3D structuring enables faster subsequent processing while maintaining high accuracy, as the difficult 2D-to-3D transformation is already completed in the input stage.
3Quantity of substance
If single-view reconstruction is used, then input data requirements are reduced, but reconstruction quality and generalization ability deteriorate
Solution Approach 1:
The patent applies dimensionality change by converting the single 2D input view into a 3D representation with added depth dimension. The depth map and 3D coordinate transformation enable the system to recover three-dimensional structure from two-dimensional input, achieving high reconstruction quality while maintaining the advantage of single-view input.
4Measurement precision
If view-specific reconstruction methods are used, then accuracy for specific views is improved, but generalization ability to arbitrary angles deteriorates
Solution Approach 1:
The patent creates a universal 3D model representation that can serve multiple viewing angles and perspectives. By reconstructing the object in three-dimensional space with depth information, the system generates a view-independent representation that can be rendered from any angle, achieving both accuracy and generalization ability for arbitrary viewing angles.
Data Source
AI summary
A method for generating a model, a terminal and a storage medium are provided. The method includes: acquiring a single first picture in which a target object is displayed; acquiring an object pose-shape parameter of the target object in the first picture; converting sampling points of the target object in the first picture from a target space to a preset canonical space according to the object pose-shape parameter; determining a global feature corresponding to the sampling points in the canonical space and a pixel-level feature corresponding to the sampling points in the canonical space; and obtaining a model parameter of the target object according to the global feature of the sampling points in the canonical space and the pixel-level feature of the sampling points in the canonical space.