Research map / Detection and segmentation / Weakly supervised detection and segmentation
Using Unreliable Pseudo-Labels for Label-Efficient Semantic Segmentation
Haochen Wang, Yuchao Wang, Yujun Shen, Junsong Fan, Yuxi Wang, Zhaoxiang Zhang
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 |
|---|---|---|---|
| use unreliable instead of reliable pixels as negative keys (with labeled) (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.35 -> 80.21 | Improved |
| add reliable pixels to unreliable negative keys (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 80.21 -> 77.57 | Hurt |
| add vanilla contrastive loss alone (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.33 -> 77.08 | Hurt |
| add category-wise memory bank, PRT and unreliable negatives to contrastive loss (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.08 -> 78.49 | Improved |
| add dynamic partition adjustment with unreliable negatives (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.08 -> 79.07 | Improved |
| add high-entropy (unreliable) filtering to the full contrastive setup (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.57 -> 79.30 | Improved |
| add prototypical denoising (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.33 -> 77.93 | Improved |
| add symmetric cross-entropy for unlabeled images (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 77.33 -> 77.69 | Improved |
| add prototypical denoising across partitions (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 79.48 -> 80.21 | Improved |
| use reverse-only CE instead of symmetric CE (ablation) | Ours · PASCAL VOC 2012 blender, 1/8 (1323) labeled | mIoU: 79.35 -> 77.76 | Hurt |
| use probability rank thresholds r_l=3, r_h=20 instead of r_l=10, r_h=20 (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 78.94 -> 80.21 | Improved |
| use initial unreliable partition 20% instead of 40% (ablation) | Ours · PASCAL VOC 2012 blender, 1/4 (2646) labeled | mIoU: 76.92 -> 80.21 | 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.
- CutMix — CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features (ICCV 2019) · 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
- Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · similarity 0.97)
- PseudoSeg: Designing Pseudo Labels for Semantic Segmentation (ICLR 2021 · similarity 0.91)
- When Confidence Fails: Revisiting Pseudo-Label Selection in Semi-supervised Semantic Segmentation (ICCV 2025 · similarity 0.91)
- Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation (ECCV 2024 · similarity 0.91)
- Active Label Correction for Semantic Segmentation with Foundation Models (ICML 2024 · similarity 0.88)
- DuPL: Dual Student with Trustworthy Progressive Learning for Robust Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.88)
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.
