Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Part Learning for Visual Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn unsupervised parts and encode with wpCOP instead of using a global descriptor | Willow actions · Willow actions | mAP: 88.5 -> 90.4 | Helped |
| Learn unsupervised parts and encode with wpCOP instead of using a global descriptor | MIT 67 · MIT 67 | accuracy: 78.5 -> 83.4 | Helped |
| Learn unsupervised parts and encode with wpCOP instead of using a global descriptor | Oxford5k, 256 dimensions · Oxford5k | mAP: 76.2 -> 84.1 | Helped |
| Learn unsupervised parts and encode with wpCOP instead of using a global descriptor | Paris6k, 64 dimensions · Paris6k | mAP: 82.5 -> 95.3 | Helped |
| Use sBOP instead of wpCOP (ablation) | Unsupervised parts on Oxford5k at 256 dimensions · Oxford5k | mAP: 84.1 -> 78.6 | Hurt |
| Use sBOP instead of wpCOP (ablation) | Unsupervised parts on Paris6k at 256 dimensions · Paris6k | mAP: 94.0 -> 90.6 | Hurt |
| Increase the number of groups from K=5 to K=20 (ablation) | Unsupervised ISA wpCOP · Willow actions | mAP: 88.4 -> 90.4 | Helped |
| Increase the number of groups from K=40 to K=80 (ablation) | Unsupervised ISA wpCOP · Willow actions | mAP: 90.3 -> 88.8 | Hurt |
| Use greedy balancing instead of iterative balancing (ablation) | Unsupervised part learning · MIT 67 | accuracy: 83.2 -> 83.4 | No clear effect |
| Use Hungarian assignment instead of iterative soft-assignment (ablation) | Unsupervised part learning · MIT 67 | accuracy: 81.2 -> 83.7; 81.3 -> 83.4 | Mixed |
| Replace VD16 Places205 initialization with VD19 initialization (ablation) | Unsupervised parts, Hungarian wpCOP, K=100 · MIT 67 | accuracy: 83.4 -> 83.5 | No clear effect |
Closest papers by meaning
- Unsupervised Part-Based Weighting Aggregation of Deep Convolutional Features for Image Retrieval (AAAI 2018 · similarity 0.89)
- Unsupervised Part Discovery from Contrastive Reconstruction (NeurIPS 2021 · similarity 0.88)
- Weakly Supervised Complementary Parts Models for Fine-Grained Image Classification From the Bottom Up (CVPR 2019 · similarity 0.88)
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · similarity 0.88)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.86)
- Adaptive Part Learning for Fine-Grained Generalized Category Discovery: A Plug-and-Play Enhancement (CVPR 2025 · similarity 0.86)
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.