Deep Common Semantic Space Embedding for Sketch-Based 3D Model Retrieval.

Area of Science:

• Computer Vision
• Machine Learning
• Computer Graphics

Background:

• Sketch-based 3D model retrieval is crucial for applications like computer graphics and computer-aided design.
• Significant visual discrepancies exist between sketches and 3D models, alongside diversity within sketch representations.
• Despite these differences, 3D models and sketches of the same object class share underlying semantic information.

Purpose of the Study:

• To develop a novel approach for sketch-based 3D model retrieval.
• To construct a deep common semantic space embedding using a triplet network.
• To address the interdomain visual perception discrepancies and intradomain diversity challenges.

Main Methods:

• Representing 3D models as a group of views to create a common data space.
• Translating 3D model views into sketches using cross-entropy evaluation to generate a common modality space.
• Learning a common semantic space embedding for sketches and 3D models via a triplet network.
• Designing four distance metrics for sketch-to-3D model comparison.

Main Results:

• The proposed method successfully achieved sketch-based 3D model retrieval.
• Experimental results on the SHREC 2013 and SHREC 2014 datasets demonstrated the method's effectiveness.
• The approach outperformed existing state-of-the-art methods in sketch-based 3D model retrieval tasks.

Conclusions:

• The developed deep common semantic space embedding effectively bridges the gap between sketch and 3D model domains.
• The novel approach significantly improves the accuracy and performance of sketch-based 3D model retrieval.
• This method offers a promising solution for efficient and accurate retrieval in computer graphics and related fields.