Research map / Detection and segmentation / Weakly supervised detection and segmentation
Learning Affinity From Attention: End-to-End Weakly-Supervised Semantic Segmentation With Transformers
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 |
|---|---|---|---|
| Use a Transformer-based end-to-end framework with learned affinity and pixel-adaptive refinement | WSSS segmentation · PASCAL VOC 2012 val | mIoU | Helped |
| Add PAR to the Transformer baseline (ablation) | WSSS model · PASCAL VOC 2012 val | mIoU: 46.7 -> 56.2 | Helped |
| Add AFA to the model with PAR (ablation) | WSSS model · PASCAL VOC 2012 val | mIoU: 56.2% to 62.6% | Helped |
| Add the auxiliary regularization loss (ablation) | WSSS model with PAR and AFA · PASCAL VOC 2012 val | mIoU: 62.6 -> 63.8 | Helped |
| Add CRF post-processing (ablation) | WSSS model with PAR, AFA, and regularization · PASCAL VOC 2012 val | mIoU: 63.8% to 66.0% | Helped |
| Learn semantic affinity from MHSA instead of omitting AFA (ablation) | Pseudo-label generation · PASCAL VOC 2012 train | mIoU: 54.4% mIoU to 66.3% mIoU | Helped |
| Add random-walk propagation with learned affinity (ablation) | AFA pseudo-label generation · PASCAL VOC 2012 train | mIoU: 66.3 -> 68.7 | Helped |
| Replace learned affinity with average vanilla MHSA for propagation (ablation) | AFA pseudo-label generation · PASCAL VOC 2012 train | mIoU: 66.3 -> 58.3 | Hurt |
| Add the position kernel to RGB-based PAR (ablation) | CAM refinement · PASCAL VOC 2012 train | mIoU: 51.7 -> 52.9 | Helped |
| Increase PAR iterations from 15 to 20 (ablation) | CAM refinement · PASCAL VOC 2012 train | mIoU: 52.9 -> 52.9 | No clear effect |
| Set affinity-loss window radius to 8 rather than 2, 4, 12, or 16 (ablation) | AFA · PASCAL VOC 2012 val | mIoU: 62.4; 62.7; 63.8; 61.5; 59.4 | Mixed |
| Use global max pooling rather than global average pooling for Transformer CAM (ablation) | CAM generation · PASCAL VOC 2012 val | mIoU: 31.1 -> 48.3 | Helped |
Closest papers by meaning
- Learning Pixel-Level Semantic Affinity With Image-Level Supervision for Weakly Supervised Semantic Segmentation (CVPR 2018 · similarity 0.93)
- Label-efficient Segmentation via Affinity Propagation (NeurIPS 2023 · similarity 0.92)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.90)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.89)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 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.