Research map / Label-efficient and robust learning / Domain adaptation and generalization
EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation
Suman Saha, Lukas Hoyer, Anton Obukhov, Dengxin Dai, Luc Van Gool
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 |
|---|---|---|---|
| share the encoder but use separate decoders instead of a fully shared network (ablation) | panoptic UDA architecture · SYNTHIA -> Cityscapes | mPQ: 34.0 -> 38.1 | Improved |
| share the encoder across tasks (M-Dec-BU vs M-Net) (ablation) | panoptic UDA architecture · SYNTHIA -> Cityscapes | mPQ: 38.1 -> 39.0 | Improved |
| use top-down instead of bottom-up instance decoder (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 39.0 -> 41.2 | Improved |
| joint training with shared encoder instead of two separate networks (M-Net) (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 38.1 -> 41.2 | Improved |
| add basic self-training on target domain (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 21.6 -> 37.5 | Improved |
| add EMA mean teacher for pseudo-labels (ablation) | EDAPS self-training · SYNTHIA -> Cityscapes | mPQ: 37.5 -> 38.9 | Improved |
| add ImageNet thing-class feature distance (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 38.9 -> 39.7 | Improved |
| add rare class sampling (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 39.7 -> 41.2 | Improved |
| use self-training instead of adversarial training (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ: 26.2 -> 41.2 | Improved |
| add instance pseudo-labels in addition to semantic pseudo-labels (ablation) | EDAPS · SYNTHIA -> Cityscapes | mPQ | No clear effect |
| add RPN losses to instance head (ablation) | EDAPS instance decoder · SYNTHIA -> Cityscapes | mPQ: 30.8 -> 41.2 | Improved |
| add RoI mask loss to instance head (ablation) | EDAPS instance decoder · SYNTHIA -> Cityscapes | mPQ: 32.7 -> 41.2 | Improved |
Later papers that compared with this method
3 later accepted papers report EDAPS as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- MC-PanDA++: Simpler, Stronger, and More Robust Domain-Adaptive Panoptic Segmentation (IJCV 2026 · better on 4 of 4 shared results)
- MC-PanDA: Mask Confidence for Panoptic Domain Adaptation (ECCV 2024 · better on 5 of 5 shared results)
- Occlusion-Aware Seamless Segmentation (ECCV 2024 · better on 3 of 4 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
- Cross-View Regularization for Domain Adaptive Panoptic Segmentation (CVPR 2021 · similarity 0.91)
- Seeing Beyond: Extrapolative Domain Adaptive Panoramic Segmentation (CVPR 2026 · similarity 0.90)
- Ada-Segment: Automated Multi-loss Adaptation for Panoptic Segmentation (AAAI 2021 · similarity 0.90)
- UniDAformer: Unified Domain Adaptive Panoptic Segmentation Transformer via Hierarchical Mask Calibration (CVPR 2023 · similarity 0.89)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.87)
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.
