Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
CLIPood: Generalizing CLIP to Out-of-Distributions
Yang Shu, Xingzhuo Guo, Jialong Wu, Ximei Wang, Jianmin Wang, Mingsheng Long
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 |
|---|---|---|---|
| Use CLIPood fine-tuning instead of CLIP ERM | CLIP ViT-B/16 · DomainBed average | accuracy: 76.7_{\pm 0.2} -> \textbf{78.6}_{\pm 0.1} | Improved |
| Use CLIPood instead of CoCoOp | CLIP ViT-B/16 · ImageNet OOD average | accuracy: 59.9 -> \textbf{60.4}_{\pm 0.1} | Improved |
| Use CLIPood rather than zero-shot CLIP | CLIP ViT-B/16 · 11-dataset open-class average | H: 71.7 -> 78.9 | Improved |
| Add adaptive MMS margins based on text-class distances (ablation) | CLIPood · DomainNet, domain shift and open classes | accuracy: 1.2% | Improved |
| Add Beta Moving Average to the fine-tuning trajectory (ablation) | CLIPood · DomainNet domain shift | accuracy: 62.3 -> 63.5; 62.7 -> 63.5 | Improved |
| Replace BMA with SWAD (ablation) | CLIPood · ImageNet open classes | H: 73.7 -> 72.7 | Hurt |
| Replace BMA with WiSE-FT (ablation) | CLIPood · SUN397 open classes | H: 80.2 -> 79.0 | Hurt |
| Use metric-softmax fine-tuning instead of a parametric linear classifier (ablation) | CLIPood · OfficeHome domain shift | accuracy | Improved |
| Set MMS lambda to 1.0, making the margin term too large (ablation) | CLIPood · DomainNet domain shift | accuracy: 63.5 -> 61.8 | Hurt |
| Use BMA beta 0.3 instead of beta 0.5 (ablation) | CLIPood · DomainNet domain shift | accuracy: 63.5 -> 63.7 | Improved |
| Use a fixed margin rather than class-adaptive margins (ablation) | CLIPood MMS · OfficeHome and DomainNet with domain shift and open classes | accuracy | Mixed |
Later papers that compared with this method
1 later accepted paper reports CLIPood as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Domain Generalization via Text-Anchored Information Bottleneck (ECCV 2026 · 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
- Does CLIP’s generalization performance mainly stem from high train-test similarity? (ICLR 2024 · similarity 0.90)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.89)
- When and How Does CLIP Enable Domain and Compositional Generalization? (ICML 2025 · similarity 0.88)
- CLIPCEIL: Domain Generalization through CLIP via Channel rEfinement and Image-text aLignment (NeurIPS 2024 · similarity 0.88)
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation (NeurIPS 2025 · similarity 0.88)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · 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.
