Research map / Detection and segmentation / Weakly supervised detection and segmentation
Token Contrast for Weakly-Supervised Semantic Segmentation
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 patch-token contrast using intermediate-layer token relations (ablation) | ViT-B WSSS · PASCAL VOC val CAM | mIoU: 27.9 to 62.5 | Helped |
| Add class-token contrast to PTC (ablation) | ViT-B WSSS · PASCAL VOC val CAM | mIoU: 62.5 to 67.2 | Helped |
| Add segmentation loss to PTC and CTC (ablation) | ViT-B WSSS · PASCAL VOC val segmentation | mIoU: 66.6 | Helped |
| Add regularization loss to the segmentation objective (ablation) | ViT-B WSSS · PASCAL VOC val segmentation | mIoU: 66.6 to 68.1 | Helped |
| Use auxiliary classifier block 10 rather than blocks 8, 9, or 11 (ablation) | ToCo · PASCAL VOC val CAM | mIoU: 63.5, 67.6, 70.5, 43.1 | Mixed |
| Use local crop size 96² instead of 64², 80², or 112² (ablation) | ToCo CTC · PASCAL VOC val segmentation | mIoU: 65.5, 67.3, 68.1, 67.0 | Mixed |
| Set EMA momentum to 0.9 instead of 0, 0.1, or 0.99 (ablation) | ToCo CTC · PASCAL VOC val segmentation | mIoU: 66.9, 67.1, 68.1, 66.6 | Mixed |
| Use absolute cosine similarity rather than signed cosine similarity in PTC (ablation) | ToCo · PASCAL VOC val CAM | mIoU: 36.5 to 70.5 | Helped |
| Use absolute cosine similarity rather than ReLU(CosSim) in PTC (ablation) | ToCo · PASCAL VOC val segmentation | mIoU: 60.1 to 68.1 | Helped |
| Use ToCo instead of ViT-PCM + CRF | ViT-B ImageNet-21k pseudo-label generation · PASCAL VOC val pseudo labels | mIoU: 69.3 to 72.3 | Helped |
| Use ToCo rather than the strongest reported single-stage baseline | Image-level WSSS · PASCAL VOC val | mIoU: 67.2 to 69.8 | Helped |
| Use ToCo rather than the strongest reported single-stage baseline | Image-level WSSS · MS COCO 2014 val | mIoU: 38.9 to 41.3 | Helped |
Closest papers by meaning
- Class Tokens Infusion for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.92)
- Weakly Supervised Semantic Segmentation via Progressive Confidence Region Expansion (CVPR 2025 · similarity 0.89)
- From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.88)
- Multi-Label Prototype Visual Spatial Search for Weakly Supervised Semantic Segmentation (CVPR 2025 · similarity 0.88)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.88)
- Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.87)
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.