Research map / Detection and segmentation / Weakly supervised detection and segmentation
Adaptive Spatial-BCE Loss for Weakly Supervised Semantic Segmentation
Tong Wu, Guangyu Gao, Junshi Huang, Xiaolin Wei, Xiaoming Wei, Chi Harold Liu
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 Spatial-BCE loss to the KL term (ablation) | pseudo-label training · PASCAL VOC 2012 train pseudo-labels (phase 1) | mIoU (%): 63.0 -> 64.2 | Improved |
| add Spatial-BCE loss to the KL term (ablation) | pseudo-label training · PASCAL VOC 2012 train pseudo-labels (phase 2) | mIoU (%): 63.7 -> 67.5 | Improved |
| determine threshold adaptively from predictions instead of fixed T/NT proportion (ablation) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels (phase 2) | mIoU (%): 60.5 -> 64.5 | Improved |
| determine threshold adaptively from predictions instead of fixed T/NT proportion (ablation) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels (phase 1) | mIoU (%): 59.7 -> 61.7 | Improved |
| iteratively update Qc for 3 phases (adaptive threshold) (ablation) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels | mIoU (%): 53.0 -> 65.3 | Improved |
| iteratively update Qc for 3 phases (best threshold) (ablation) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels | mIoU (%): 53.0 -> 68.1 | Improved |
| add dCRF to pseudo-labels after 3 phases (best threshold) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels | mIoU (%): 68.1 -> 70.4 | Improved |
| add dCRF to pseudo-labels after 3 phases (adaptive threshold) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels | mIoU (%): 65.3 -> 66.3 | Improved |
| add Qc update phase 1 to phase 3 (adaptive threshold) (ablation) | Spatial-BCE · PASCAL VOC 2012 train pseudo-labels | mIoU (%): 61.7 -> 64.5 -> 65.3 | Improved |
| use saliency maps as post-processing in addition to image labels | Spatial-BCE · PASCAL VOC 2012 val | mIoU (%): 70.0 -> 71.8 | Improved |
| use saliency maps as post-processing in addition to image labels | Spatial-BCE · MS-COCO 2014 val | mIoU (%): 35.2 -> 38.6 | Improved |
| use Spatial-BCE pseudo-labels with best threshold instead of adaptive | Spatial-BCE · PASCAL VOC 2012 val | mIoU (%): 68.5 -> 70.0 | Improved |
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.
- ECS-Net — ECS-Net: Improving Weakly Supervised Semantic Segmentation by Using Connections Between Class Activation Maps (ICCV 2021) · better on 1 of 1 shared result
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
- Memory-Based Cross-Image Contexts for Weakly Supervised Semantic Segmentation (TPAMI 2022 · similarity 0.91)
- Class Re-Activation Maps for Weakly-Supervised Semantic Segmentation (CVPR 2022 · similarity 0.89)
- Looking Beyond Single Images for Weakly Supervised Semantic Segmentation Learning (TPAMI 2022 · similarity 0.89)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.89)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.87)
- Multi-Label Prototype Visual Spatial Search for Weakly Supervised Semantic Segmentation (CVPR 2025 · 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.
