Research map / Label-efficient and robust learning / Domain adaptation and generalization
Zero-Shot Domain Generalization via Domain Prompt Learning
Shiming Chen, Salman Khan, Salman Khan, Fahad Shahbaz Khan
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 visual domain learning (VDL) on top of text domain learning (ablation) | DoP · DG average (4 datasets, ViT-B/16) | accuracy: 75.8 -> 82.0 | Improved |
| add text domain learning (TDL) on top of visual domain learning (ablation) | DoP · DG average (4 datasets, ViT-B/16) | accuracy: 77.1 -> 82.0 | Improved |
| add style token selection (STS) in TDL (ablation) | DoP · DG average (4 datasets, ViT-B/16) | accuracy: 81.4 -> 82.0 | Improved |
| prompt with unlearned 'A random style of a [class]' (DP) instead of standard prompt | ZS-CLIP · ZSDG average (ViT-B/16) | accuracy: 79.6 -> 76.3 | Hurt |
| prompt with unlearned random style (DP) instead of standard prompt | ZS-CLIP · ZSDG average (ViT-L/14) | accuracy: 83.0 -> 80.0 | Hurt |
| prompt with unlearned random style (DP) instead of standard prompt | ZS-CLIP · ZSDG average (ResNet-50) | accuracy: 70.2 -> 72.7 | Improved |
| use DoP instead of PromptStyler (selection plus visual refinement) | PromptStyler · DG average (ViT-B/16) | accuracy: 80.8 -> 82.0 | Improved |
| use DoP instead of PromptStyler | PromptStyler · DG average (ResNet-50) | accuracy: 74.7 -> 74.9 | Improved |
| use DoP instead of PromptStyler on PACS Sketch | PromptStyler · PACS Sketch (ViT-B/16) | accuracy: 92.3 -> 91.5 | Hurt |
| use DoP instead of TPT with ViT-L/14 | TPT · ZSDG average (ViT-L/14) | accuracy: 84.1 -> 83.3 | Hurt |
| refine only the selected style token instead of many prompt parameters | TPT · PACS per-image test time (ms) | time: 462.8 -> 170.5 | Improved |
| vary number of style tokens K too small or too large (ablation) | DoP · PACS / OfficeHome | accuracy | Mixed |
Closest papers by meaning
- Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · similarity 0.94)
- OSLoPrompt: Bridging Low-Supervision Challenges and Open-Set Domain Generalization in CLIP (CVPR 2025 · similarity 0.92)
- Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · similarity 0.91)
- CLIPCEIL: Domain Generalization through CLIP via Channel rEfinement and Image-text aLignment (NeurIPS 2024 · similarity 0.91)
- PracticalDG: Perturbation Distillation on Vision-Language Models for Hybrid Domain Generalization (CVPR 2024 · similarity 0.91)
- Multimodal Unsupervised Domain Generalization by Retrieving Across the Modality Gap (ICLR 2025 · similarity 0.91)
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.
