Research map / Detection and segmentation / Open-set and few-shot detection
Self-Supervised Learning of Object Parts for Semantic Segmentation
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 |
|---|---|---|---|
| Fine-tune with Leopart on ImageNet+COCO rather than ImageNet alone (ablation) | Leopart · PVOC12 | mIoU (LC): 68.0 -> 69.3 | Helped |
| Focus clustering on foreground tokens rather than all tokens (ablation) | Leopart · PVOC12 val | mIoU (K=500): 47.8 -> 50.7 | Helped |
| Focus clustering on foreground tokens rather than background tokens (ablation) | Leopart · PVOC12 val | mIoU (K=500): 41.4 -> 50.7 | Helped |
| Add local crops to the two global crops (ablation) | Leopart · PVOC12 val | mIoU (K=500): 45.0 -> 50.7 | Helped |
| Increase local crops from two to four (ablation) | Leopart · PVOC12 val | mIoU (K=500): 49.3 -> 50.7 | Helped |
| Use an EMA teacher (ablation) | Leopart · PVOC12 val | mIoU (K=500): 47.9 -> 50.7 | Helped |
| Use 300 prototypes rather than 100 (ablation) | Leopart · PVOC12 val | mIoU (K=500): 49.2 -> 50.7 | Helped |
| Use 300 prototypes rather than 500 (ablation) | Leopart · PVOC12 val | mIoU (K=500): 49.1 -> 50.7 | Helped |
| Add a clustering queue (ablation) | Leopart · PVOC12 val | mIoU (K=500): 48.1 -> 50.7 | Helped |
| Add Leopart fine-tuning to DINO (ablation) | DINO · PVOC12 val | mIoU (fully unsupervised): 4.6 -> 18.9 (+14.3%) | Helped |
| Add cluster-based foreground extraction (CBFE) (ablation) | Leopart · PVOC12 val | mIoU (fully unsupervised): 18.9 -> 36.6 (+17.7%) | Helped |
| Add community detection (CD) (ablation) | Leopart + CBFE · PVOC12 val | mIoU (fully unsupervised): 36.6 -> 41.7 (+5.1%) | Helped |
Closest papers by meaning
- Unsupervised Semantic Segmentation with Self-supervised Object-centric Representations (ICLR 2023 · similarity 0.93)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.90)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.90)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.89)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.89)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.88)
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.