Research map / Label-efficient and robust learning / Test-time adaptation
Towards Purified Multi-Label Test-Time Adaptation of Vision-Language Models
Yiwen Liang, Hui Chen, Yizhe Xiong, Mengyao Lyu, Yuhan Cao, Zijia Lin, SHUAICHENG NIU, Sicheng Zhao, Jungong Han, Guiguang Ding
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 region purification (RP) (ablation) | PuRF baseline · COCO2017 (ViT-B/16) | mAP (%): 60.71 -> 63.35 | Improved |
| add episodic purification (EP) of cache (ablation) | PuRF baseline · COCO2017 (ViT-B/16) | mAP (%): 60.71 -> 62.59 | Improved |
| add episodic purification to region purification (ablation) | PuRF (RP only) · COCO2017 (ViT-B/16) | mAP (%): 63.35 -> 64.56 | Improved |
| add temporal refreshing (TR) to RP+EP (ablation) | PuRF (RP+EP) · COCO2017 (ViT-B/16) | mAP (%): 64.56 -> 65.35 | Improved |
| add region purification to EP+TR (ablation) | PuRF (EP+TR) · COCO2017 (ViT-B/16) | mAP (%): 63.37 -> 65.35 | Improved |
| select regions by relative score instead of using all regions (ablation) | PuRF · COCO2017 (ViT-B/16) | mAP (%): 63.46 -> 65.35 | Improved |
| relative-score instead of absolute-score region selection (ablation) | PuRF · COCO2017 (ViT-B/16) | mAP (%): 65.02 -> 65.35 | Improved |
| use purified-region cache with entropy filtering instead of global-only cache (ablation) | PuRF · COCO2017 (ViT-B/16) | mAP (%): 62.21 -> 65.35 | Improved |
| max aggregation instead of average aggregation of cached regions (ablation) | PuRF · COCO2017 (ViT-B/16) | mAP (%): 64.92 -> 65.35 | Improved |
| min-entropy cache retrieval instead of global retrieval (ablation) | PuRF · COCO2017 (ViT-B/16) | mAP (%): 64.80 -> 65.35 | Improved |
| PuRF instead of ReTA on ViT-B/32 | ReTA · Average of five datasets (ViT-B/32) | mAP (%): 65.83 -> 69.88 | Improved |
| PuRF instead of ReTA with MaPLe init | ReTA · Average of five datasets (ViT-B/16, MaPLe init) | mAP (%): 70.88 -> 73.79 | Improved |
Closest papers by meaning
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.89)
- Multi-Label Test-Time Adaptation with Bound Entropy Minimization (ICLR 2025 · similarity 0.88)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- Bayesian Test-Time Adaptation for Vision-Language Models (CVPR 2025 · similarity 0.87)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 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.
