Research map / Label-efficient and robust learning / Semi-supervised learning
Weighting Pseudo-Labels via High-Activation Feature Index Similarity and Object Detection for Semi-Supervised Segmentation
Prantik Howlader, Hieu Le, Dimitris Samaras
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 per-pixel rank-statistics pseudo-label weighting plus detector-based reliable pixels to existing SSL frameworks | UniMatch, AugSeg, AEL, U2PL · Pascal VOC Classic 1/16 | mIoU: AEL 66.1 -> 69.8 (ResNet-101) | Improved |
| Add the method to the same four frameworks on Cityscapes | UniMatch, AugSeg, AEL, U2PL · Cityscapes val 1/16 | mIoU: AEL 74.5 -> 77.9 (ResNet-101) | Improved |
| Add pseudo-label pixel weighting (prototypes from labeled pixels only) (ablation) | basic teacher-student baseline · Pascal VOC Classic 1/16 | mIoU: +3.9% | Improved |
| Add object-detector-based reliable pixel identification on top of weighting (ablation) | basic teacher-student baseline · Pascal VOC Classic 1/16 | mIoU: +3.9% -> +7.4% | Improved |
| Use rank statistics instead of cosine similarity for weights | UniMatch + Ours · Pascal VOC pseudo-label accuracy | pseudo-label accuracy | Improved |
| Agreement between detector and segmentation model to select pixels (ablation) | DeepLabv3+ R101 pseudo-labels · Pascal VOC 1/16 | pseudo-label accuracy: Cat 32.8 -> 41.7 | Improved |
| Vary k in top-k rank statistics | Ours · Pascal VOC 1/16, 1/8 | mIoU | Mixed |
| Raise detector confidence threshold from 0.85 to 0.95 (ablation) | Ours (UniMatch) · Pascal VOC Classic 1/16 | mIoU: 76.7 -> 75.7 | Hurt |
| Increase memory bank size from 32 to 256 per class (ablation) | Ours (UniMatch) · Pascal VOC Classic 1/16 | mIoU: 74.6 -> 76.7 | Improved |
| Set unsupervised loss weight alpha to 0.4 (ablation) | Ours (UniMatch) · Pascal VOC Classic 1/16 | mIoU: 45.1 -> 76.7 (alpha 0 vs 0.4) | Improved |
| Apply the weighting to a ViT-based method (SemiVL) | SemiVL · Cityscapes 1/16 | mIoU: 77.9 -> 79.1 | Improved |
| Apply the weighting to UniMatch on MS COCO | UniMatch (Xception-65) · MS COCO 1/512 | mIoU: 31.9 -> 33.7 | 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.
- SemiVL — SemiVL: Semi-Supervised Semantic Segmentation with Vision-Language Guidance (ECCV 2024) · 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 0.92)
- Querying Labeled for Unlabeled: Cross-Image Semantic Consistency Guided Semi-Supervised Semantic Segmentation (TPAMI 2023 · similarity 0.92)
- PseudoSeg: Designing Pseudo Labels for Semantic Segmentation (ICLR 2021 · similarity 0.92)
- DAW: Exploring the Better Weighting Function for Semi-supervised Semantic Segmentation (NeurIPS 2023 · similarity 0.92)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · similarity 0.90)
- 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.
