Research map / Detection and segmentation / Weakly supervised detection and segmentation
Box2Seg: Attention Weighted Loss and Discriminative Feature Learning for Weakly Supervised Segmentation
Viveka Kulharia, Siddhartha Chandra, Amit Agrawal, Philip Torr, Ambrish Tyagi
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 GrabCut pseudo-labels instead of box masks only (ablation) | segmentation network · PASCAL VOC 2012 val | mIoU: 59.3 -> 72.7 | Improved |
| add affinity (pixel embedding) loss (ablation) | GrabCut baseline · PASCAL VOC 2012 val | mIoU: 72.7 -> 73.9 | Improved |
| add attention weighted loss with box regularization (ablation) | GrabCut baseline · PASCAL VOC 2012 val | mIoU: 72.7 -> 74.1 | Improved |
| add attention weighted loss with fill-rate regularization (ablation) | GrabCut baseline · PASCAL VOC 2012 val | mIoU: 72.7 -> 74.6 | Improved |
| use fill-rate regularization instead of box regularization of attention map (ablation) | attention weighted loss · PASCAL VOC 2012 val | mIoU: 74.1 -> 74.6 | Improved |
| add affinity loss on top of attention weighted loss (ablation) | AW-fr · PASCAL VOC 2012 val | mIoU: 74.6 -> 74.9 | Improved |
| add attention weighted loss in fully supervised training (ablation) | supervised baseline · PASCAL VOC 2012 val (fully supervised) | mIoU: 73.6 -> 75.1 | Improved |
| add dense CRF post-processing (ablation) | Box2Seg · PASCAL VOC 2012 val | mIoU: 74.9 -> 76.4 | Improved |
| fine-tune with 5% pixel-level labels | Box2Seg · PASCAL VOC 2012 val (semi-supervised) | mIoU: 74.9 -> 78.7 | Improved |
| fine-tune with 10% pixel-level labels | Box2Seg · PASCAL VOC 2012 val (semi-supervised) | mIoU: 74.9 -> 83.1 | Improved |
| fine-tune with 100% pixel-level labels | Box2Seg · PASCAL VOC 2012 val (semi-supervised) | mIoU: 74.9 -> 86.4 | Improved |
| use Box2Seg weak model as initialization instead of ImageNet | fully supervised DeepLab · PASCAL VOC 2012 val (semi-supervised) | mIoU: 74.5 -> 86.4 | Improved |
Later papers that compared with this method
3 later accepted papers report Box2Seg as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- [Re] Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (NeurIPS 2022 · better on 0 of 1 shared result)
- Affinity Attention Graph Neural Network for Weakly Supervised Semantic Segmentation (TPAMI 2021 · better on 2 of 2 shared results)
- Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (CVPR 2021 · better on 0 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
- Weakly Supervised Semantic Segmentation via Box-Driven Masking and Filling Rate Shifting (TPAMI 2023 · similarity 0.92)
- Box-Driven Class-Wise Region Masking and Filling Rate Guided Loss for Weakly Supervised Semantic Segmentation (CVPR 2019 · similarity 0.92)
- Simple Does It: Weakly Supervised Instance and Semantic Segmentation (CVPR 2017 · similarity 0.91)
- Background-Aware Pooling and Noise-Aware Loss for Weakly-Supervised Semantic Segmentation (CVPR 2021 · similarity 0.91)
- Reliability Does Matter: An End-to-End Weakly Supervised Semantic Segmentation Approach (AAAI 2020 · similarity 0.91)
- BoxTeacher: Exploring High-Quality Pseudo Labels for Weakly Supervised Instance Segmentation (CVPR 2023 · similarity 0.90)
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.
