Research map / 3D vision / Face and shape
Info3D: Representation Learning on 3D Objects using Mutual Information Maximization and Contrastive Learning
Aditya Sanghi
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 |
|---|---|---|---|
| use embeddings from layer 6 instead of layer 7 on aligned data (ablation) | Info3D PointNet · ModelNet40 aligned clustering | AMI: 0.401 ± 0.005 -> 0.633 ± 0.002 | Improved |
| use embeddings from layer 7 instead of layer 6 on unaligned data (ablation) | Info3D PointNet · ModelNet40 unaligned clustering | AMI: 0.404 ± 0.002 -> 0.496 ± 0.004 | Improved |
| use cosine-distance chunk as second view instead of translation (ablation) | Info3D PointNet · ModelNet40 aligned clustering | AMI: 0.633 ± 0.002 -> 0.677 ± 0.002 | Improved |
| use cosine chunk rather than axis chopped chunk (ablation) | Info3D PointNet · ModelNet40 aligned clustering | AMI: 0.660 ± 0.002 -> 0.677 ± 0.002 | Improved |
| use axis chopped chunk instead of translation (ablation) | Info3D PointNet · ModelNet40 transfer | accuracy: 87.8 -> 89.8 | Improved |
| use axis chopped chunk rather than cosine distance chunk (ablation) | Info3D DGCNN · ModelNet40 transfer | accuracy: 90.8 -> 91.6 | Improved |
| replace PointNet with DGCNN encoder (ablation) | Info3D axis chopped chunk · ModelNet40 transfer | accuracy: 89.8 -> 91.6 | Improved |
| add translation to rotation SO(3) + uniform scale (ablation) | Info3D unaligned · ModelNet40 unaligned clustering | AMI: 0.313 ± 0.005 -> 0.500 ± 0.001 | Improved |
| add uniform scale to rotation SO(3) + translate (ablation) | Info3D unaligned · ModelNet40 unaligned clustering | AMI: 0.496 ± 0.004 -> 0.500 ± 0.001 | No clear effect |
| add random scale to rotation SO(3) + translate (ablation) | Info3D unaligned · ModelNet40 unaligned clustering | AMI: 0.496 ± 0.004 -> 0.485 ± 0.004 | No clear effect |
| chunk size 512 instead of 128 (ablation) | Info3D chunks · ModelNet40 aligned clustering | AMI: 0.653 ± 0.002 -> 0.677 ± 0.002 | Improved |
| chunk size 512 instead of 768 (ablation) | Info3D chunks · ModelNet40 aligned clustering | AMI: 0.658 ± 0.008 -> 0.677 ± 0.002 | Improved |
Closest papers by meaning
- Single Image 3D Shape Retrieval via Cross-Modal Instance and Category Contrastive Learning (ICCV 2021 · similarity 0.87)
- Unsupervised Contrastive Learning for Efficient and Robust Spectral Shape Matching (AAAI 2026 · similarity 0.87)
- Toward Realistic Single-View 3D Object Reconstruction With Unsupervised Learning From Multiple Images (ICCV 2021 · similarity 0.87)
- Multiview Aggregation for Learning Category-Specific Shape Reconstruction (NeurIPS 2019 · similarity 0.87)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.86)
- Duoduo CLIP: Efficient 3D Understanding with Multi-View Images (ICLR 2025 · similarity 0.85)
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.
