Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain Adaptation with a Single Vision-Language Embedding
Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette
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 |
|---|---|---|---|
| fine-tune classifier on PIN-augmented features from a single prompt | source-only model · ACDC Night (CS source) | mIoU (%): 18.31 -> 25.03 | Improved |
| use PODA instead of CLIPstyler stylized images | source-only model · Cityscapes (GTA5 source) | mIoU (%): 31.50 -> 40.08 | Improved |
| CLIPstyler stylized images for adaptation | source-only model · ACDC Rain (CS source) | mIoU (%): 38.20 -> 37.17 | Hurt |
| freeze CLIP backbone during source training (ablation) | CLIP-RN50 segmenter · ACDC Snow (CS source) | mIoU (%): 22.27 -> 39.28 | Improved |
| augment Layer2/3/4 features in addition to Layer1 (ablation) | PODA · ACDC Night (CS source) | mIoU (%): 25.03 -> 23.43, 22.93, 21.05 | Hurt |
| increase number of mined styles from 1 to 2975 (ablation) | PODA · ACDC Night (CS source) | mIoU (%): 16.00 -> 25.03 | Improved |
| use semantically unrelated random prompts (ablation) | PODA · ACDC Night (CS source) | mIoU (%): 25.03 -> 20.45 | Hurt |
| paraphrase prompt with ChatGPT (relevant prompts) (ablation) | PODA · ACDC Night (CS source) | mIoU (%): 25.03 -> 24.82 | No clear effect |
| add optimized concept S* to style text (ablation) | PODA · ACDC Snow (CS source) | mIoU (%): 43.90 -> 44.72 | Improved |
| use style text only without content word driving (ablation) | PODA · ACDC Snow (CS source) | mIoU (%): 43.90 -> 40.45 | Hurt |
| initialize style stats from zeros/ones instead of source statistics (ablation) | PODA · ACDC Night (CS source) | mIoU (%): 25.03 -> 8.59 | Hurt |
| skip source-only pretraining, train from scratch on augmented features (ablation) | PODA · ACDC Snow (CS source) | mIoU (%): 43.90 -> 36.73 | Hurt |
Closest papers by meaning
- Domain-Agnostic Mutual Prompting for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.89)
- Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation (CVPR 2025 · similarity 0.88)
- Approximate Domain Unlearning for Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.88)
- Unified Language-driven Zero-shot Domain Adaptation (CVPR 2024 · similarity 0.87)
- Leveraging Vision-Language Models for Improving Domain Generalization in Image Classification (CVPR 2024 · 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.
