Research map / Label-efficient and robust learning / Domain adaptation and generalization
CoDA: Instructive Chain-of-Domain Adaptation with Severity-Aware Visual Prompt Tuning
ZiYang Gong, FuHao Li, Yupeng Deng, Deblina Bhattacharjee, Xianzheng Ma, Xiangwei Zhu, Zhenming Ji
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 Chain-of-Domain (CoD) staged adaptation plus traditional sampling (ablation) | MIC · Cityscapes to ACDC val | mIoU: 68.3 -> 70.3 | Improved |
| Add CoD+traditional sampling to reduce seed variance (ablation) | MIC · Cityscapes to ACDC val, 3 seeds | mIoU range: 4.7 -> 0.7 | Improved |
| Add CoD only (no traditional sampling) to DAFormer (ablation) | DAFormer · Cityscapes to ACDC val, 3 seeds | mIoU range: 1.7 -> 1.7 | No clear effect |
| Add SAVPT without CoD (ablation) | MIC · Cityscapes to ACDC val | mIoU: 68.3 -> 64.3 | Hurt |
| Add SAVPT on top of CoD+traditional and generated images X_G (ablation) | MIC with CoD+Tra.+X_G · Cityscapes to ACDC val | mIoU: 70.4 -> 72.1 | Improved |
| Add severity perception trigger (SPT) to visual prompts (ablation) | CoDA · Cityscapes to ACDC val | mIoU: 67.5 -> 72.1 | Improved |
| Add generated mini-dataset X_G to intermediate domain (ablation) | MIC with CoD+Tra. · Cityscapes to ACDC val | mIoU: 70.3 -> 70.4 | Improved |
| Activate SAVPT at inference time (ablation) | MIC with CoDA · Cityscapes to ACDC val | mIoU: 72.6 (w/o) vs 72.5 (w/) | No clear effect |
| Initialize Meta-Visual Prompts at zero instead of ones (ablation) | SAVPT · Cityscapes to ACDC val | mIoU: 67.7 (delta=1, Corner+Center) -> 72.1 (delta=0) | Improved |
| Uniform vs normal initialization of prompts (ablation) | SAVPT · Cityscapes to ACDC val | mIoU: 69.7 vs 69.6 (Corner+Center) | No clear effect |
| Use severity threshold 0.05 instead of other values (ablation) | CoDA · Nighttime Driving val | mIoU: 56 (0.03) -> 59.2 (0.05) | Improved |
| Use four separate intermediate domains instead of two (rain+fog+snow, night) (ablation) | CoD · Cityscapes to ACDC val | mIoU: 65.4-67.1 (four domains) vs 70.3 (two domains) | 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.
- HRDA — HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (ECCV 2022) · better on 4 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
- Domain-Agnostic Mutual Prompting for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.88)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.88)
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.88)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Exploring Sparse Visual Prompt for Domain Adaptive Dense Prediction (AAAI 2024 · similarity 0.87)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.86)
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.
