Research map / Detection and segmentation / Weakly supervised detection and segmentation
Discriminative Region Suppression 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 |
|---|---|---|---|
| Suppress discriminative feature activations with DRS rather than leaving the classifier unmodified (ablation) | VGG-16 classification network with DeepLab-Large-FOV segmentation · PASCAL VOC 2012 val | mIoU: 50.1% -> 62.9% | Helped |
| Apply DRS in higher-level layers rather than only lower-level layers (ablation) | VGG-16 classification network with DeepLab-Large-FOV segmentation · PASCAL VOC 2012 val | mIoU: 53.6% -> 62.9% | Helped |
| Add DRS to lower-level layers when higher-level layers already use it (ablation) | VGG-16 classification network with DeepLab-Large-FOV segmentation · PASCAL VOC 2012 val | mIoU: 62.7% -> 62.9% | No clear effect |
| Use a learnable controller instead of a fixed controller with δ=0.55 (ablation) | DRS classification network · PASCAL VOC 2012 val | mIoU: 62.8% -> 62.9% | No clear effect |
| Reduce the fixed suppression threshold from δ=0.90 to δ=0.55 (ablation) | DRS with a non-learnable controller · PASCAL VOC 2012 val | mIoU: 51.9% -> 62.8% | Helped |
| Over-suppress with a fixed threshold of δ=0.40 rather than use δ=0.55 (ablation) | DRS with a non-learnable controller · PASCAL VOC 2012 val | mIoU: 62.8% -> 59.6% | Hurt |
| Add localization map refinement learning (ablation) | DRS with a learnable controller · PASCAL VOC 2012 val | mIoU: 62.9% -> 63.5% | Helped |
| Add localization map refinement learning (ablation) | DRS with a non-learnable controller (δ=0.55) · PASCAL VOC 2012 val | mIoU: 62.8% -> 63.6% | Helped |
| Use DRS with non-learnable controller and refinement learning | DeepLab-ASPP with ResNet-101 · PASCAL VOC 2012 test | mIoU: 67.2% -> 71.4% | Helped |
| Use DRS with learnable controller and refinement learning | DeepLab-Large-FOV with ResNet-101 · PASCAL VOC 2012 test | mIoU: 68.0% -> 67.5% | Hurt |
Closest papers by meaning
- Embedded Discriminative Attention Mechanism for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.90)
- Threshold Matters in WSSS: Manipulating the Activation for the Robust and Accurate Segmentation Model Against Thresholds (CVPR 2022 · similarity 0.89)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Weakly Supervised Semantic Segmentation via Progressive Confidence Region Expansion (CVPR 2025 · similarity 0.88)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.88)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 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.