Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly Supervised Semantic Segmentation by Pixel-to-Prototype Contrast
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 |
|---|---|---|---|
| Add pixel-to-prototype contrast | SEAM · PASCAL VOC 2012 val, DeepLab-LargeFOV | mIoU (%): 64.5 -> 67.7 | Helped |
| Add pixel-to-prototype contrast | EPS · PASCAL VOC 2012 test, DeepLab-ASPP | mIoU (%): 70.8 -> 73.6 | Helped |
| Add cross-prototype contrast (ablation) | SEAM · PASCAL VOC 2012 train | mIoU (%): 55.41 -> 59.08 | Helped |
| Add cross-CAM contrast (ablation) | SEAM with cross-prototype contrast · PASCAL VOC 2012 train | mIoU (%): 59.08 -> 60.35 | Helped |
| Add intra-view contrast without mining strategies (ablation) | SEAM with cross-view contrast · PASCAL VOC 2012 train and val | mIoU (%) | Hurt |
| Add semi-hard prototype mining (ablation) | SEAM with intra-view contrast · PASCAL VOC 2012 train and val | mIoU (%): +1.74% and 1.44% | Helped |
| Add hard pixel sampling (ablation) | SEAM with intra-view contrast · PASCAL VOC 2012 train and val | mIoU (%): +1.74% and 1.44% | Helped |
| Set prototype top K to 32 (ablation) | Pixel-to-prototype contrast with SEAM · PASCAL VOC 2012 train | mIoU (%): 55.41 -> 61.54 | Helped |
| Vary prototype top K from 4 to 64 (ablation) | Pixel-to-prototype contrast with SEAM · PASCAL VOC 2012 train | mIoU (%): 60.02; 60.87; 61.08; 61.54; 60.55 | Mixed |
| Add horizontal flip to rescaling (ablation) | Pixel-to-prototype contrast with SEAM · PASCAL VOC 2012 train | mIoU (%): 61.54 -> 61.63 | No clear effect |
| Use rotation with rescaling (ablation) | Pixel-to-prototype contrast with SEAM · PASCAL VOC 2012 train | mIoU (%): 61.54 -> 58.57 | Hurt |
| Use translation with rescaling (ablation) | Pixel-to-prototype contrast with SEAM · PASCAL VOC 2012 train | mIoU (%): 61.54 -> 59.31 | Hurt |
Closest papers by meaning
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.89)
- POT: Prototypical Optimal Transport for Weakly Supervised Semantic Segmentation (CVPR 2025 · similarity 0.88)
- [Re] Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (NeurIPS 2022 · similarity 0.87)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.87)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.85)
- Toward Modality Gap: Vision Prototype Learning for Weakly-supervised Semantic Segmentation with CLIP (AAAI 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.