Research map / Label-efficient and robust learning / Domain adaptation and generalization
Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 weak-label-guided prototype computation instead of plain feature averaging (ablation) | WDASS prototypes · GTA5->Cityscapes | mIoU: +2.5 image, +1.6 point, +2.3 coarse | Helped |
| Asymmetric inter-domain alignment (source features to target prototypes) instead of symmetric alignment (ablation) | Contrastive alignment · GTA5->Cityscapes | mIoU: +2.9 image, +1.2 point, +1.7 coarse | Helped |
| Pixel-prototype contrastive alignment instead of pixel-pixel alignment (ablation) | Contrastive alignment · GTA5->Cityscapes | mIoU: +5.1 image, +2.0 point, +2.1 coarse | Helped |
| Add image loss (L_image) to the base weak-label training (ablation) | Image-label WDASS · GTA5->Cityscapes | mIoU: 36.6 to 56.8 mIoU (image label) | Helped |
| Add boundary loss to base (ablation) | WDASS with image, point, coarse labels · GTA5->Cityscapes | mIoU: +5.9 image, +3.9 point (mean-based from base), +2.6 coarse | Helped |
| Add intra-domain contrastive loss (L_intra) to point-label model (ablation) | Contrastive alignment · GTA5->Cityscapes | mIoU: +3.6 mIoU point | Helped |
| Add inter-domain contrastive loss (L_inter) to point-label model (ablation) | Contrastive alignment · GTA5->Cityscapes | mIoU: +2.3 mIoU point | Helped |
| Ours with coarse labels vs supervised training on full labels | Coarse-label WDASS · GTA5->Cityscapes | mIoU: +1.7 over supervised (69.1 vs 67.4) | Helped |
| Use weak target labels with the proposed framework instead of UDA | Image, point, coarse label WDASS · GTA5->Cityscapes | mIoU: image 61.5, point 64.7 vs UDA SoTA 57.5 | Helped |
Closest papers by meaning
- Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · similarity 0.91)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.89)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.89)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.89)
- Toward Modality Gap: Vision Prototype Learning for Weakly-supervised Semantic Segmentation with CLIP (AAAI 2025 · similarity 0.89)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · 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.