3D Mesh Retrieval Using Self-Supervised Geometric Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D object retrieval methods face challenges in efficiently scaling up due to the effort involved in annotation, particularly in supervised methods, and view-based approaches lose dimensionality, leading to data loss.
Innovation Solution
A self-supervised method using a neural network model to learn effective embeddings of 3D mesh objects by pre-processing query objects into a unit sphere with a reduced number of triangles, representing each triangle as a 2D matrix, and employing a self-supervised network to embed historical models in an embedding space for geometric similarity retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised methods are used for 3D object retrieval, then retrieval accuracy can be improved, but the effort involved in annotation increases significantly
Solution Approach 1:
The system employs self-supervised learning where the model learns to embed 3D objects by automatically generating supervision signals from the data itself, eliminating the need for manual annotation. The neural network learns geometric features and relationships through self-supervised tasks on raw 3D mesh data, allowing the system to serve itself without external annotation effort while maintaining retrieval accuracy
2Ease of manufacture
If view-based methods are used for 3D retrieval, then implementation simplicity is improved, but dimensionality information is lost
Solution Approach 1:
The system transitions from 2D view-based representations to 3D mesh-based representations by embedding objects in a geometric embedding space that preserves three-dimensional structural information. The neural network processes 3D mesh data directly, maintaining depth, volume, and spatial relationships that are lost in 2D projections, thereby recovering dimensional information while keeping the approach simpler than traditional 3D methods
3Adaptability or versatility
If large numbers of 3D models are stored in databases, then object variety is improved, but retrieval efficiency deteriorates
Solution Approach 1:
The system replaces traditional mechanical search methods (direct comparison of 3D models) with a neural network-based embedding system. The neural network learns geometric feature representations and stores them in an embedding space, allowing for efficient similarity search through distance metrics in the embedding space rather than computationally expensive 3D model comparisons, thereby maintaining retrieval efficiency despite large database sizes
Data Source
AI summary
A system and processor-implemented method for three-dimensional (3D) object retrieval using a self-supervised model is provided. The present system learns an embedding space of the 3D mesh objects in a self-supervised manner without the need for objects annotated with their class or other properties. Effective embeddings of 3D mesh objects are learned using the self-supervised method for ranked retrieval from a large collection of 3D objects. A simple representation of mesh objects and a standard neural network model is used to learn the embedding. The results are retrieved on the basis of the shape of the object which may not belong to the same category but look similar in shape using the embeddings generated by self-supervised model. The system is independent of class labels and uses the entire 3D model for better information extraction.