Research map / Detection and segmentation / Weakly supervised detection and segmentation
Know Your Attention Maps: Class-specific Token Masking for Weakly Supervised Semantic Segmentation
Joëlle Hanna, Damian Borth
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 random masking of non-label [CLS] tokens (ablation) | Multi-[CLS] ViT-B WSSS · MS COCO pseudo-masks | mIoU: 41.9 -> 43.2 | Improved |
| add random masking of non-label [CLS] tokens (ablation) | Multi-[CLS] ViT-B WSSS · VOC pseudo-masks | mIoU: 71.6 -> 72.7 | Improved |
| add [REG] register token (ablation) | Multi-[CLS] ViT-B WSSS · MS COCO pseudo-masks | mIoU: 42.8 -> 43.2 | Improved |
| add [REG] register token (ablation) | Multi-[CLS] ViT-B WSSS · VOC pseudo-masks | mIoU: 72.3 -> 72.7 | Improved |
| add attention head pruning (ablation) | Multi-[CLS] ViT-B WSSS · MS COCO pseudo-masks | mIoU: 41.7 -> 43.2 | Improved |
| add attention head pruning (ablation) | Multi-[CLS] ViT-B WSSS · VOC pseudo-masks | mIoU: 72.0 -> 72.7 | Improved |
| raise [CLS] masking ratio from 0% towards 50% (ablation) | Multi-[CLS] ViT-B WSSS · DFC2020/Endotect/ADE20K pseudo-masks | mIoU: performance improves up to around 50% then plateaus or slightly declines | Mixed |
| use multiple class-specific [CLS] ViT instead of Sparse ViT | Multi-[CLS] ViT-B WSSS · DFC2020 (specialized, segmentation) | mIoU: 52.2 -> 67.2 | Improved |
| use multiple class-specific [CLS] ViT instead of Sparse ViT | Multi-[CLS] ViT-B WSSS · ADE20K (specialized, segmentation) | mIoU: 34.5 -> 38.2 | Improved |
| use multiple class-specific [CLS] ViT instead of Sparse ViT | Multi-[CLS] ViT-B WSSS · Endotect (specialized, segmentation) | mIoU: 68.0 -> 69.8 | Improved |
| proposed method versus DuPL on VOC val segmentation | Multi-[CLS] ViT-B WSSS · COCO val (final segmentation) | mIoU: 43.6 -> 43.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DuPL — DuPL: Dual Student with Trustworthy Progressive Learning for Robust Weakly Supervised Semantic Segmentation (CVPR 2024) · better on 3 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Pseudo-Mask Matters in Weakly-Supervised Semantic Segmentation (ICCV 2021 · similarity 0.92)
- Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.91)
- DHR: Dual Features-Driven Hierarchical Rebalancing in Inter- and Intra-Class Regions for Weakly-Supervised Semantic Segmentation (ECCV 2024 · similarity 0.91)
- Class Tokens Infusion for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.90)
- Weakly Supervised Semantic Segmentation via Progressive Confidence Region Expansion (CVPR 2025 · similarity 0.89)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.89)
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.
