Research map / Label-efficient and robust learning / Semi-supervised learning
From Softmax to Dirichlet: Evidential Learning for Semi-supervised Semantic Segmentation
Huayu Mai, Rui Sun, Yujia Chen, Wangkai Li, Bingzhou Wang, Aibing Li, Zhangyu He, Yuan Wang
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 |
|---|---|---|---|
| replace softmax-score filtering with Dirichlet evidential uncertainty (ESS) (ablation) | UniMatch · Pascal classic 1/16, ResNet-101 | mIoU: 75.2 -> 76.9 | Improved |
| add hyper-evidence (collective evidence) on top of ESS (ablation) | UniMatch + ESS · Pascal classic 1/16, ResNet-101 | mIoU: 76.9 -> 77.5 | Improved |
| replace softmax-score filtering with Dirichlet evidential uncertainty (ESS) (ablation) | UniMatch · Pascal classic 1/4, ResNet-101 | mIoU: 78.8 -> 79.5 | Improved |
| add hyper-evidence on top of ESS (ablation) | UniMatch + ESS · Pascal classic 1/4, ResNet-101 | mIoU: 79.5 -> 80.0 | Improved |
| integrate HESS into UniMatch | UniMatch · Pascal high-quality 1/16, ResNet-101 | mIoU: 75.2 -> 77.5 | Improved |
| integrate HESS into UniMatch V2 | UniMatch V2 · Pascal high-quality 1/16, DINOv2-S | mIoU: 79.0 -> 80.9 | Improved |
| integrate HESS into UniMatch | UniMatch · Cityscapes 1/16, ResNet-101 | mIoU: 76.7 -> 78.0 | Improved |
| integrate HESS into UniMatch V2 | UniMatch V2 · Cityscapes 1/16, DINOv2-S | mIoU: 80.6 -> 81.8 | Improved |
| integrate HESS into UniMatch | UniMatch · COCO 1/512, XC-65 | mIoU: 31.9 -> 33.8 | Improved |
| integrate HESS into UniMatch V2 | UniMatch V2 · COCO 1/512, DINOv2-S | mIoU: 39.3 -> 41.4 | Improved |
| raise softmax threshold from 0.95 to 0.99 (ablation) | UniMatch softmax filtering · Pascal classic 1/16, ResNet-101 | mIoU: 75.2 -> 73.9 | Hurt |
| set uncertainty threshold too low (0.10) or too high (0.40) instead of 0.25 (ablation) | HESS uncertainty filtering · Pascal classic 1/16, ResNet-101 | mIoU: 77.5 -> 74.1 (0.10), 74.5 (0.40) | Hurt |
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.
- UniMatch — UniMatch: Universal Matching from Atom to Task for Few-Shot Drug Discovery (ICLR 2025) · better on 10 of 10 shared results
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
- When Confidence Fails: Revisiting Pseudo-Label Selection in Semi-supervised Semantic Segmentation (ICCV 2025 · similarity 0.89)
- Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · similarity 0.88)
- Boosting Semi-Supervised Semantic Segmentation with Probabilistic Representations (AAAI 2023 · similarity 0.88)
- Multi-view Evidential Learning-based Medical Image Segmentation (AAAI 2025 · similarity 0.86)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · similarity 0.86)
- UncertainSAM: Fast and Efficient Uncertainty Quantification of the Segment Anything Model (ICML 2025 · 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.
