Research map / Label-efficient and robust learning / Domain adaptation and generalization
Behind Every Domain There is a Shift: Adapting Distortion-Aware Vision Transformers for Panoramic Semantic Segmentation
Jiaming Zhang, Kailun Yang, Hao Shi, Simon Reiß, Kunyu Peng, Chaoxiang Ma, Haodong Fu, Philip H. S. Torr, Kaiwei Wang, Rainer Stiefelhagen
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 deformable patch embedding (DPE) (ablation) | Trans4PASS+ baseline · DensePASS (Cityscapes 512x512) | mIoU: 39.02 -> 43.11 | Improved |
| add DMLPv2 on top of DPE (ablation) | Trans4PASS+ · DensePASS (Cityscapes 512x512) | mIoU: 43.11 -> 49.94 | Improved |
| use DMLPv2 instead of DMLPv1 with DPE (ablation) | Trans4PASS+ · DensePASS (Cityscapes 512x512) | mIoU: 45.89 -> 49.94 | Improved |
| add DMLPv1 on top of DPE (ablation) | Trans4PASS+ · DensePASS (Cityscapes 512x512) | mIoU: 43.11 -> 45.89 | Improved |
| add DMLPv2 (channel mixer) alongside pooling mixer (ablation) | DMLPv2 · DensePASS (Cityscapes 512x512) | mIoU: 46.66 -> 49.94 | Improved |
| add pooling mixer alongside channel mixer (ablation) | DMLPv2 · DensePASS (Cityscapes 512x512) | mIoU: 47.66 -> 49.94 | Improved |
| use DMLPv2 instead of PoolFormer+FAN token mixing (ablation) | Trans4PASS+ · DensePASS (Cityscapes 512x512) | mIoU: 45.97 -> 49.94 | Improved |
| use DMLPv2 instead of CycleMLP (ablation) | Trans4PASS+ with DPE · DensePASS (Cityscapes 512x512) | mIoU: 40.16 -> 49.94 | Improved |
| use DPE instead of vanilla patch embedding with DMLPv2 (ablation) | Trans4PASS+ · DensePASS (Cityscapes 512x512) | mIoU: 47.08 -> 49.94 | Improved |
| use DMLPv2 instead of vanilla MLP: Cityscapes accuracy cost (ablation) | Trans4PASS+ · Cityscapes val (Cityscapes 512x512) | mIoU: 74.93 -> 65.86 | Hurt |
| replace SSL pseudo labels with SAM-rectified labels plus MPA | Trans4PASS+ (S) · DensePASS UDA | mIoU: 56.38 -> 59.43 | Improved |
| upgrade Trans4PASS to Trans4PASS+ (Small) | Trans4PASS · SynPASS test | mIoU: 38.57 -> 40.72 | Improved |
Closest papers by meaning
- Bending Reality: Distortion-Aware Transformers for Adapting to Panoramic Semantic Segmentation (CVPR 2022 · similarity 0.95)
- Semantics Distortion and Style Matter: Towards Source-free UDA for Panoramic Segmentation (CVPR 2024 · similarity 0.91)
- Both Style and Distortion Matter: Dual-Path Unsupervised Domain Adaptation for Panoramic Semantic Segmentation (CVPR 2023 · similarity 0.91)
- GoodSAM: Bridging Domain and Capacity Gaps via Segment Anything Model for Distortion-aware Panoramic Semantic Segmentation (CVPR 2024 · similarity 0.90)
- Geometric Exploitation for Indoor Panoramic Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
- AdapToPASS: Ambiguity-aware Adaptive Spherical Transformer for Panoramic Semantic Segmentation (NeurIPS 2026 · 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.
