Research map / Vision-language and multimodal / CLIP and zero-shot
Condensed Test-Time Adaptation of VLMs for Action Recognition
Wenxuan Ge, Hongyu Qu, Rui Yan, Guo-Sen Xie, Yazhou Yao, Xiangbo Shu, Jinhui Tang
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 PSPS (probabilistic semantic patch selection) (ablation) | CONDA / ViCLIP · Kinetics-600 | top-1 accuracy: 70.9 -> 71.2 | Improved |
| add ATC (adaptive tube construction) (ablation) | CONDA / ViCLIP · COIN | top-1 accuracy: 65.6 -> 66.4 | Improved |
| combine PSPS and ATC (ablation) | CONDA / ViCLIP · COIN | top-1 accuracy: 65.6 -> 67.4 | Improved |
| use probabilistic selection instead of top-k hard selection (ablation) | CONDA · Kinetics-600 | top-1 accuracy: 71.6 -> 72.1 | Improved |
| use both gradients and attention maps in semantic patch activation (ablation) | CONDA · COIN | top-1 accuracy: 66.7 -> 67.4 | Improved |
| add spatial expansion in tubes (ablation) | CONDA · COIN | top-1 accuracy: 66.3 -> 67.4 | Improved |
| use random temporal expansion (ablation) | CONDA · Kinetics-600 | top-1 accuracy: 71.6 -> 71.5 | No clear effect |
| use adaptive instead of fixed temporal expansion (ablation) | CONDA · COIN | top-1 accuracy: 66.9 -> 67.4 | Improved |
| vary number of selected patches k from 5 to 20 (ablation) | CONDA · Kinetics-600 | top-1 accuracy: 71.8 -> 72.1 | No clear effect |
| add CONDA to TDA-style cache on ViCLIP | ViCLIP · Kinetics-600 | top-1 accuracy: 70.9 -> 72.1 | Improved |
| add CONDA to vanilla CLIP ViT-L/14 | CLIP ViT-L/14 · Kinetics-600 | top-1 accuracy: 76.4 -> 76.9 | Improved |
| use CONDA instead of DPE on egocentric EGTEA | EgoVLP · EGTEA | top-1 accuracy: 19.7 -> 22.7 | Improved |
Closest papers by meaning
- Modality-Collaborative Test-Time Adaptation for Action Recognition (CVPR 2024 · similarity 0.89)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Dual Prototype Evolving for Test-Time Generalization of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · similarity 0.88)
- Efficient Transfer Learning for Video-language Foundation Models (CVPR 2025 · similarity 0.88)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · 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.
