Research map / Detection and segmentation / Weakly supervised detection and segmentation
MARS: Model-agnostic Biased Object Removal without Additional Supervision for Weakly-Supervised Semantic Segmentation
Sanghyun Jo, In-Jae Yu, Kyungsu Kim
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 MARS biased-object removal on top of RS+EPM | RS+EPM · PASCAL VOC 2012 val | mIoU: 74.4 -> 77.7 | Improved |
| add MARS to IRNet | IRNet · PASCAL VOC 2012 val | mIoU: 63.5 -> 69.8 | Improved |
| add MARS to SEAM instead of ADELE | SEAM · PASCAL VOC 2012 val | mIoU: 69.3 (ADELE) -> 70.8 | Improved |
| add MARS to AdvCAM instead of W-OoD (which uses an extra dataset) | AdvCAM · PASCAL VOC 2012 val | mIoU: 69.8 (W-OoD) -> 70.3 | Improved |
| use STEGO instead of Leopart as the unsupervised segmentation model (ablation) | MARS with RS+EPM · PASCAL VOC 2012 val | mIoU: 75.4 -> 77.7 | Improved |
| add complementing of debiased labels with online model predictions (ablation) | MARS · PASCAL VOC 2012 train | mIoU: 77.4 -> 80.9 | Improved |
| add WCE (certain-mask weighting) to complementing (ablation) | MARS complementing · PASCAL VOC 2012 train | mIoU: 80.9 -> 81.8 | Improved |
| add WCE to complementing, effect on false negatives (ablation) | MARS complementing · PASCAL VOC 2012 train | FN: 0.075 -> 0.090 | Hurt |
| vary K_bg from 1 to 5 (ablation) | MARS debiased centroid selection · PASCAL VOC 2012 val | mIoU | No clear effect |
| use larger alpha (> 0.5) in centroid selection (ablation) | MARS · PASCAL VOC 2012 val | mIoU | Hurt |
| apply MARS to classes with poor initial pseudo masks (e.g., tv/monitor) | RS+EPM · PASCAL VOC 2012 val per-class | IoU: tv 60.2 -> 57.3; train 56.2 -> 85.3 | Mixed |
Closest papers by meaning
- Re-Distributing Biased Pseudo Labels for Semi-Supervised Semantic Segmentation: A Baseline Investigation (ICCV 2021 · similarity 0.87)
- ModuSeg: Decoupling Object Discovery and Semantic Retrieval for Training-Free Weakly Supervised Segmentation (ECCV 2026 · similarity 0.87)
- Semi-supervised Semantic Segmentation via Strong-weak Dual-branch Network (ECCV 2020 · similarity 0.87)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · similarity 0.86)
- Frozen CLIP: A Strong Backbone for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 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.
