Research map / Label-efficient and robust learning / Semi-supervised learning
Universal Semi-Supervised Semantic Segmentation
Tarun Kalluri, Girish Varma, Manmohan Chandraker, C.V. Jawahar
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 within- and cross-dataset entropy losses to joint training (ablation) | Univ-basic · CS+CamVid average (N=100, ResNet-18) | mIoU: 43.97 -> 47.83 | Improved |
| add within- and cross-dataset entropy losses to joint training (ablation) | Univ-basic · CS+CamVid average (N=50, ResNet-18) | mIoU: 40.69 -> 43.62 | Improved |
| add cross-dataset entropy loss only (ablation) | Univ-basic · CamVid test (N=100, ResNet-18) | mIoU: 51.90 -> 49.31 | Hurt |
| add within-dataset entropy loss on top of cross-dataset loss (ablation) | Univ-cross · CS+CamVid average (N=100, ResNet-18) | mIoU: 43.57 -> 47.83 | Improved |
| add entropy losses to joint training (ablation) | Univ-basic · CS+IDD average (N=100, ResNet-18) | mIoU: 31.58 -> 31.97 | Improved |
| add entropy losses to joint training (ablation) | Univ-basic · CS+IDD average (N=1500, ResNet-50) | mIoU: 58.38 -> 59.63 | Improved |
| add entropy losses to joint training (ablation) | Univ-basic · CS+SUN-RGBD average (1.5k labeled, ResNet-50) | mIoU: 44.78 -> 50.52 | Improved |
| add entropy losses to joint training (ablation) | Univ-basic · Cityscapes val (N=100, CS+IDD, ResNet-18) | mIoU: 37.94 -> 36.48 | Hurt |
| use K=3 prototype embeddings per label instead of 1 (ablation) | Univ-full · CS+CamVid average (N=100, ResNet-18) | mIoU: 47.83 -> 48.39 | Improved |
| use K=5 prototype embeddings per label instead of 1 (ablation) | Univ-full · CS+CamVid average (N=100, ResNet-18) | mIoU: 47.83 -> 47.55 | Hurt |
| use prototype label embeddings instead of Word2Vec (ablation) | Univ-full · CS+CamVid average (N=100, ResNet-18) | mIoU: 44.45 -> 47.83 | Improved |
| keep label embeddings fixed instead of updating them (theta=0.5) (ablation) | Univ-full · CS+CamVid average (N=100, ResNet-18) | mIoU: 41.50 -> 47.83 | Improved |
Closest papers by meaning
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Improving Semi-Supervised and Domain-Adaptive Semantic Segmentation with Self-Supervised Depth Estimation (IJCV 2023 · similarity 0.91)
- Semi-supervised Semantic Segmentation with Prototype-based Consistency Regularization (NeurIPS 2022 · similarity 0.91)
- AllSpark: Reborn Labeled Features from Unlabeled in Transformer for Semi-Supervised Semantic Segmentation (CVPR 2024 · similarity 0.91)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · similarity 0.90)
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.
