Research map / 3D vision / Face and shape
Single Image 3D Shape Retrieval via Cross-Modal Instance and Category Contrastive Learning
Ming-Xian Lin, Jie Yang, He Wang, Yu-Kun Lai, Rongfei Jia, Binqiang Zhao, Lin Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add category-level contrastive loss (ablation) | Ours · Pix3D | AccTop-1: 75.2% -> 78.9% | Improved |
| add category-level contrastive loss (ablation) | Ours · Pix3D | AccCats: 95.9% -> 96.1% | No clear effect |
| use color transfer instead of only affine augmentation (ablation) | Ours · Pix3D | AccTop-1: 75.2% -> 78.9% | Improved |
| use color transfer instead of only affine augmentation (ablation) | Ours · Comp Cars | AccTop-1: 67.3% -> 77.8% | Improved |
| use color transfer instead of only affine augmentation (ablation) | Ours · Stanford Cars | AccTop-1: 81.7% -> 83.4% | Improved |
| use color transfer instead of HSV augmentation (ablation) | Ours · Pix3D | AccTop-1: 76.6% -> 78.9% | Improved |
| use color transfer instead of HSV augmentation (ablation) | Ours · Comp Cars | AccTop-1: 73.0% -> 77.8% | Improved |
| use HSV augmentation instead of only affine augmentation | Ours · Stanford Cars | AccTop-1: 81.7% -> 81.5% | Hurt |
| use contrastive instance loss instead of triplet loss with category supervision (ablation) | Ours · Pix3D | AccTop-1: 66.5% -> 78.9% | Improved |
| use contrastive instance loss instead of triplet loss (ablation) | Ours · Comp Cars | AccTop-1: 64.0% -> 77.8% | Improved |
| use contrastive instance loss instead of triplet loss (ablation) | Ours · Stanford Cars | AccTop-1: 80.3% -> 83.4% | Improved |
| include ground-truth shape in category loss positives (ablation) | Ours · Pix3D | AccTop-1: 76.7% -> 78.9% | Improved |
Closest papers by meaning
- Hard Example Generation by Texture Synthesis for Cross-domain Shape Similarity Learning (NeurIPS 2020 · similarity 0.92)
- Angular Triplet-Center Loss for Multi-View 3D Shape Retrieval (AAAI 2019 · similarity 0.88)
- Doodle to Object: Practical Zero-Shot Sketch-Based 3D Shape Retrieval (AAAI 2023 · similarity 0.88)
- Enhancing 2D Representation via Adjacent Views for 3D Shape Retrieval (ICCV 2019 · similarity 0.88)
- Info3D: Representation Learning on 3D Objects using Mutual Information Maximization and Contrastive Learning (ECCV 2020 · similarity 0.87)
- Shape2Scene: 3D Scene Representation Learning Through Pre-training on Shape Data (ECCV 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
