Research map / Label-efficient and robust learning / Test-time adaptation
Test-Time Distribution Normalization for Contrastively Learned Visual-language Models
Yifei Zhou, Juntao Ren, Fengyu Li, Ramin Zabih, Ser-Nam Lim
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 distribution normalization (DN) | TCL · ImageNet1K zero-shot | Acc@1: 20.0 -> 30.5 | Improved |
| add DN | CLIP · MSCOCO image-to-text | R@1: 52.4 -> 51.7 | No clear effect |
| add DN | CLIP · MSCOCO text-to-image | R@1: 30.2 -> 33.4 | Improved |
| add DN on top of test-time augmentation (TTA) | CLIP + TTA · cross-modal retrieval | average recall: 1.3% average improvement | Improved |
| average DN with vanilla dot product (DN*) instead of DN | CLIP + TTA + DN · retrieval | std of improvement: 1.3% -> 0.8% | Improved |
| add DN to CLIP | CLIP · Flickr8k-Expert | Kendall tau_c: 51.4 -> 54.3 | Improved |
| add DN to TCL | TCL · Flickr8k-Expert | Kendall tau_c: 31.0 -> 42.0 | Improved |
| add DN to ALBEF | ALBEF · Flickr8k-Expert | Kendall tau_c: 24.9 -> 34.8 | Improved |
| add DN to ALBEF | ALBEF · ImageNet1K zero-shot | Acc@1: 37.3 -> 39.9 | Improved |
| add DN to ALBEF | ALBEF · Pascal-50S | mean accuracy: 76.2 -> 75.6 | Hurt |
| add DN to TCL | TCL · Pascal-50S | mean accuracy: 65.6 -> 70.3 | Improved |
| estimate mean from 1 training sample instead of 100 (DN*) (ablation) | CLIP + DN* · ImageNet1K zero-shot | Acc@1: 61.7 -> 60.6 | Hurt |
Closest papers by meaning
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation (NeurIPS 2025 · similarity 0.89)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.89)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.88)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.88)
- CLIPLoss and Norm-Based Data Selection Methods for Multimodal Contrastive Learning (NeurIPS 2024 · similarity 0.87)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 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.
