Research map / Label-efficient and robust learning / Semi-supervised learning
Re-Distributing Biased Pseudo Labels for Semi-Supervised Semantic Segmentation: A Baseline Investigation
Ruifei He, Jihan Yang, Xiaojuan Qi
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 random sampling after distribution alignment (DARS vs DA+TS) (ablation) | DA + TS · Cityscapes 1/8, crop 361 | mIoU: 67.31 -> 68.01 | Improved |
| replace single thresholding (ST) with DARS pseudo-labeling | self-training · Cityscapes 1/8, crop 361 | mIoU: 65.70 -> 68.01 | Improved |
| replace class-balanced thresholding (CBST) with DARS | self-training · Cityscapes 1/8, crop 361 | mIoU: 66.29 -> 68.01 | Improved |
| add temperature scaling to ST (ablation) | ST · Cityscapes 1/8, crop 361 | mIoU: 65.70 -> 65.89 | Improved |
| add temperature scaling to CBST (ablation) | CBST · Cityscapes 1/8, crop 361 | mIoU: 66.29 -> 66.61 | Improved |
| add iterative training (2 rounds) to DARS (ablation) | DARS · Cityscapes 1/8, crop 361 | mIoU: 68.01 -> 69.64 | Improved |
| add iterative training to CBST (ablation) | CBST · Cityscapes 1/8, crop 361 | mIoU: 66.29 -> 67.20 | Improved |
| add stronger augmentation (0.2, 0.5) at round 1 without enlarging labeling ratio (ablation) | DARS round 1 · Cityscapes 1/8, crop 361 | mIoU: 68.01 -> 67.54 | Hurt |
| add second self-training round with alpha 20% (ablation) | DARS · Cityscapes 1/8, crop 361 | mIoU: 68.01 -> 68.27 | Improved |
| enlarge labeling ratio to alpha 50% at round 2 (ablation) | DARS · Cityscapes 1/8, crop 361 | mIoU: 68.27 -> 68.93 | Improved |
| add stronger random scaling range (beta 0.2, 0.5) at round 2 with alpha 50% (ablation) | DARS · Cityscapes 1/8, crop 361 | mIoU: 68.93 -> 69.64 | Improved |
| use alpha 60% instead of 50% at round 2 (ablation) | DARS · Cityscapes 1/8, crop 361 | mIoU: 68.93 -> 68.75 | Hurt |
Closest papers by meaning
- PseudoSeg: Designing Pseudo Labels for Semantic Segmentation (ICLR 2021 · similarity 0.92)
- Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · similarity 0.92)
- DAW: Exploring the Better Weighting Function for Semi-supervised Semantic Segmentation (NeurIPS 2023 · similarity 0.92)
- Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation (ECCV 2024 · similarity 0.92)
- Learning from Future: A Novel Self-Training Framework for Semantic Segmentation (NeurIPS 2022 · similarity 0.91)
- RWMS: Reliable Weighted Multi-Phase for Semi-supervised Segmentation (AAAI 2024 · similarity 0.89)
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.
