Research map / Detection and segmentation / Weakly supervised detection and segmentation
Reducing Information Bottleneck 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 |
|---|---|---|---|
| Use RIB in place of IRN | weakly supervised semantic segmentation · MS COCO 2014 val | mIoU: 41.4 -> 43.8 | Helped |
| Use RIB in place of the strongest image-label baseline | weakly supervised semantic segmentation · PASCAL VOC 2012 val | mIoU: 66.3 -> 68.3 | Helped |
| Add salient object supervision | RIB · PASCAL VOC 2012 val | mIoU: 68.3 -> 70.2 | Helped |
| Replace RIB loss with BCE using a double-sided saturating activation (ablation) | RIB fine-tuning · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 49.7; 50.5; 50.9 | Hurt |
| Replace GNDRP with GAP (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 54.8 | Hurt |
| Set GNDRP threshold tau to 0.3 or 0.5 instead of 0.4 (ablation) | RIB-GNDRP · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 55.8; 56.0 | No clear effect |
| Reduce margin m to 300 (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 54.0 | Hurt |
| Use fine-tuning learning rate 5 x 10^-6 instead of 8 x 10^-6 (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 54.9 | Hurt |
| Use fine-tuning learning rate 1 x 10^-5 instead of 8 x 10^-6 (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 56.0 | Hurt |
| Increase margin m to 1000 (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 55.9 | Hurt |
| Use only the target image in the RIB adaptation batch (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 55.3 | Hurt |
| Increase RIB batch size from 20 to 30 (ablation) | RIB · PASCAL VOC 2012 train, initial seed | mIoU: 56.5 -> 56.7 | No clear effect |
Closest papers by meaning
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.90)
- Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (CVPR 2021 · similarity 0.89)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Less Is More: Label Recommendation for Weakly Supervised Point Cloud Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Separate and Conquer: Decoupling Co-occurrence via Decomposition and Representation for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.86)
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.