Research map / Label-efficient and robust learning / Test-time adaptation
Learning Instance-Specific Adaptation for Cross-Domain Segmentation
Yuliang Zou, Zizhao Zhang, Chun-Liang Li, Han Zhang, Tomas Pfister, Jia-Bin Huang
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 |
|---|---|---|---|
| learn per-channel vector calibration strength instead of one scalar (ablation) | InstCal · GTA5 to Cityscapes | mIoU: 39.8 -> 41.1 | Improved |
| learn separate calibration vectors for mean and variance (ablation) | InstCal-U · GTA5 to Cityscapes | mIoU: 41.1 -> 41.5 | Improved |
| learn calibration strength instead of fixed m=0.1 (ablation) | InstCal-U · GTA5 to Cityscapes | mIoU: 40.1 -> 41.5 | Improved |
| train calibration with DeepAugment instead of default weak augmentation (ablation) | InstCal-U · GTA5 to Cityscapes | mIoU: 39.7 -> 41.5 | Improved |
| use RandAugment instead of DeepAugment for calibration training (ablation) | InstCal-C · GTA5 to Cityscapes | mIoU: 42.2 -> 40.0 | Hurt |
| pre-train with DeepAugment but no InstCal (ablation) | pre-trained DeepLabv2 · GTA5 to Cityscapes | mIoU: 35.7 -> 31.7 | Hurt |
| use conditional (MLP-weighted basis) calibration instead of unconditional | InstCal-U · GTA5 to Cityscapes | mIoU: 41.5 -> 42.2 (GTA5 avg 40.3 -> 41.1); Cityscapes-source avg 45.9 -> 45.0 | Mixed |
| use 8 calibration basis instead of 16 for conditional calibration (ablation) | InstCal-C · GTA5 to Cityscapes | mIoU: 40.7 -> 42.2 | Improved |
| use per-instance statistics instead of batch statistics (batch size 16) (ablation) | InstCal-U · GTA5 to Cityscapes | mIoU: 40.7 -> 41.5 | Improved |
| add test-time entropy minimization | InstCal-U · GTA5 to Cityscapes | mIoU: 41.5 -> 44.1 | Improved |
| add test-time entropy minimization | InstCal-C · GTA5 to Cityscapes | mIoU: 42.2 -> 44.2 | Improved |
| add InstCal-U to non-DG pre-trained model | DeepLabv3+ ResNet-50 · GTA5 to 3 real domains (avg) | mIoU: 27.9 -> 37.1 | Improved |
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.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 3 of 3 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
- Adapt Before Comparison: A New Perspective on Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.89)
- CrDoCo: Pixel-Level Domain Transfer With Cross-Domain Consistency (CVPR 2019 · similarity 0.88)
- Domain-Agnostic Prior for Transfer Semantic Segmentation (CVPR 2022 · similarity 0.88)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.87)
- Domain-Rectifying Adapter for Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.87)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 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.
