Research map / Detection and segmentation / Weakly supervised detection and segmentation
Unsupervised Semantic Segmentation with Self-supervised Object-centric Representations
Read the paperCheck 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 |
|---|---|---|---|
| Add one self-training iteration to the initial pseudo-masks (ablation) | COMUS · PASCAL VOC 2012 val | mIoU: 43.8 $\pm$ 0.1 -> 47.6 $\pm$ 0.4 | Helped |
| Add a second self-training iteration on the larger trainaug set (ablation) | COMUS · PASCAL VOC 2012 val | mIoU: 47.6 $\pm$ 0.4 -> 50.0 $\pm$ 0.4 | Helped |
| Add self-training to the initial pseudo-masks (ablation) | COMUS · MS COCO val (81 categories) | mIoU: 18.2 -> 19.6 | Helped |
| Train COMUS on PASCAL VOC before evaluating on MS COCO (ablation) | COMUS · MS COCO val (21 PASCAL VOC classes) | mIoU: 28.8 -> 31.9 | Helped |
| Replace self-supervised BasNet saliency masks with Spectral Decomposition masks (ablation) | COMUS (Iteration 1) · PASCAL VOC 2012 val | mIoU: 47.6 -> 43.8 | Hurt |
| Replace self-supervised BasNet saliency masks with DeepUSPS masks (ablation) | COMUS (Iteration 1) · PASCAL VOC 2012 val | mIoU: 47.6 -> 45.5 | Hurt |
| Use supervised instead of unsupervised saliency masks (ablation) | COMUS (Iteration 2) · PASCAL VOC 2012 val | mIoU: 50.0 -> 52.3 | Helped |
| Use spectral clustering with a Laplacian eigenmap instead of k-means clustering (ablation) | COMUS · PASCAL VOC | mIoU | Helped |
| Filter the most distant, uncertain cluster samples (ablation) | COMUS · PASCAL VOC | mIoU | Helped |
| Use more than 40% filtered examples (ablation) | COMUS · PASCAL VOC | mIoU | Hurt |
| Use more than two self-training iterations on the same dataset (ablation) | COMUS · PASCAL VOC trainaug | mIoU | Hurt |
| Increase the cluster count for self-training pseudo-masks (ablation) | COMUS · PASCAL VOC over-clustering | mIoU | Hurt |
Closest papers by meaning
- Self-Supervised Learning of Object Parts for Semantic Segmentation (CVPR 2022 · similarity 0.93)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.90)
- Unsupervised Universal Image Segmentation (CVPR 2024 · similarity 0.89)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.88)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.88)
- CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers (CVPR 2024 · 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.