Research map / Label-efficient and robust learning / Domain adaptation and generalization
Look at the Neighbor: Distortion-aware Unsupervised Domain Adaptation for Panoramic Semantic Segmentation
Xu Zheng, Tianbo Pan, Yunhao Luo, Lin 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 |
|---|---|---|---|
| add distortion-aware attention in the deepest layer (after ESA layers) (ablation) | DATR-M · Cityscapes-to-DensePASS (source only) | mIoU: 36.45 -> 38.48 | Improved |
| add DA in the shallow layers (ablation) | DATR-M · Cityscapes-to-DensePASS (source only) | mIoU: 36.45 -> 37.56 | No clear effect |
| use relative positional encoding instead of absolute positional encoding (ablation) | DATR-M · SynPASS-to-DensePASS13 | mIoU: 41.56 -> 51.04 | Improved |
| add relative positional encoding versus none (ablation) | DATR-M · SynPASS-to-DensePASS13 | mIoU: 36.98 -> 51.04 | Improved |
| add class-wise feature aggregation on top of pseudo-label self-training (ablation) | DATR-M · Cityscapes-to-DensePASS | mIoU: 45.71 -> 52.90 | Improved |
| add CFA on top of self-training (ablation) | DATR-S · SynPASS-to-DensePASS13 | mIoU: 50.08 -> 54.05 | Improved |
| replace MPA with CFA (ablation) | DATR-M/T/S · SynPASS-to-DensePASS13 | mIoU: 48.24 -> 51.04 | Improved |
| replace non-iterative or linear class-center update with iterative update (ablation) | CFA · SynPASS-to-DensePASS13 (DATR-M) | mIoU: 50.43 / 50.53 -> 51.04 | Improved |
| use DATR backbones instead of PVT-S under MPA | MPA · SynPASS-to-DensePASS13 | mIoU: 40.90 -> 52.76 | Improved |
| compact DATR-M versus Trans4PASS-S backbone without adaptation on Cityscapes-trained models | DATR-M · Cityscapes (CS) mIoU | mIoU on Cityscapes: 81.17 -> 75.23 | Hurt |
| change neighborhood region size of DA away from 11x11 | DATR-M · DensePASS test | 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.
- DAFormer — DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (CVPR 2022) · 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
- Both Style and Distortion Matter: Dual-Path Unsupervised Domain Adaptation for Panoramic Semantic Segmentation (CVPR 2023 · similarity 0.95)
- Semantics Distortion and Style Matter: Towards Source-free UDA for Panoramic Segmentation (CVPR 2024 · similarity 0.94)
- GoodSAM: Bridging Domain and Capacity Gaps via Segment Anything Model for Distortion-aware Panoramic Semantic Segmentation (CVPR 2024 · similarity 0.93)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.88)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.87)
- AdapToPASS: Ambiguity-aware Adaptive Spherical Transformer for Panoramic Semantic Segmentation (NeurIPS 2026 · 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.
