Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain Adaptive Semantic Segmentation Using Weak Labels
Sujoy Paul, Yi-Hsuan Tsai, Samuel Schulter, Amit K. Roy-Chowdhury, Manmohan Chandraker
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 weak-label classification loss L_c (pseudo-weak) to output-space alignment baseline (ablation) | AdaptOutput baseline · GTA5 -> Cityscapes (no pixel-level adaptation) | mIoU (%): 41.4 -> 44.2 | Improved |
| add category-wise alignment L_adv on top of L_c (pseudo-weak) (ablation) | Ours (UDA) · GTA5 -> Cityscapes (no pixel-level adaptation) | mIoU (%): 44.2 -> 45.6 | Improved |
| add category-wise alignment L_adv on top of L_c (oracle-weak) (ablation) | Ours (WDA) · GTA5 -> Cityscapes (no pixel-level adaptation) | mIoU (%): 50.8 -> 52.1 | Improved |
| add category-wise alignment L_adv on top of L_c (pseudo-weak) (ablation) | Ours (UDA) · SYNTHIA -> Cityscapes (16 classes) (no pixel-level adaptation) | mIoU (%): 41.7 -> 42.7 | Improved |
| add category-wise alignment L_adv on top of L_c (oracle-weak) (ablation) | Ours (WDA) · SYNTHIA -> Cityscapes (16 classes) (no pixel-level adaptation) | mIoU (%): 47.8 -> 49.2 | Improved |
| add pixel-level adaptation to full pseudo-weak method (ablation) | Ours (UDA) · GTA5 -> Cityscapes | mIoU (%): 45.6 -> 48.2 | Improved |
| add pixel-level adaptation to full oracle-weak method (ablation) | Ours (WDA: Image) · GTA5 -> Cityscapes | mIoU (%): 52.1 -> 53.0 | Improved |
| use oracle-weak labels instead of pseudo-weak labels | Ours · GTA5 -> Cityscapes (no pixel-level adaptation) | mIoU (%): 45.6 -> 52.1 | Improved |
| use point supervision instead of image-level oracle-weak labels | Ours (WDA) · GTA5 -> Cityscapes | mIoU (%): 53.0 -> 56.4 | Improved |
| use point supervision instead of image-level oracle-weak labels | Ours (WDA) · SYNTHIA -> Cityscapes (16 classes) | mIoU (%): 50.6 -> 57.2 | Improved |
| use 5-point supervision instead of image-level labels | Ours (WDA) · GTA5 -> Cityscapes | mIoU (%): 59.4 vs 65.1 fully supervised | Improved |
| raise pseudo-weak label threshold T above 0.3 (ablation) | Ours (UDA) · GTA5 -> Cityscapes | mIoU (%) | 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.
- DADA — DADA: Depth-Aware Domain Adaptation in Semantic Segmentation (ICCV 2019) · 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
- Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning (CVPR 2023 · similarity 0.94)
- LabOR: Labeling Only if Required for Domain Adaptive Semantic Segmentation (ICCV 2021 · similarity 0.92)
- Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · similarity 0.92)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.92)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.91)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · similarity 0.91)
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.
