Research map / Video / Action recognition
Expanding Language-Image Pretrained Models for General Video Recognition
Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling
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 cross-frame communication to CLIP-Mean (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +1.2% | Improved |
| add 1-layer multi-frame integration transformer (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +0.5% | Improved |
| add video-specific prompting (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +2.3% over CLIP-Mean | Improved |
| add multi-view inference (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +1.5% | Improved |
| finetune the image encoder instead of freezing it (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +3.0% | Improved |
| freeze the text encoder (ablation) | X-CLIP-B/16 8f · Kinetics-400 | CUDA memory: 22G -> 6G | Improved |
| add text branch instead of random FC head (ablation) | X-CLIP-B/16 8f · HMDB-51 few-shot K=2 | top-1 accuracy: +18.8% | Improved |
| add text branch instead of random FC head (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: +0.7% | Improved |
| CLIP pretraining instead of ImageNet-21k (ablation) | X-CLIP-B/16 8f · Kinetics-400 | top-1 accuracy: 79.8 -> 82.3 | Improved |
| video-specific prompts instead of handcrafted ensemble / learnable vectors (ablation) | X-CLIP-B/16 · UCF-101 zero-shot | top-1 accuracy: +6.1% | Improved |
| sparse instead of dense sampling (ablation) | X-CLIP-B/16 · Kinetics-400 | top-1 accuracy | Improved |
| cross-frame attention vs CLIP-Joint space-time attention (ablation) | X-CLIP-B/16 · Kinetics-400 fully-supervised / HMDB-51 few-shot | top-1 accuracy: +0.2% and +9.5% | Improved |
Closest papers by meaning
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.89)
- OST: Refining Text Knowledge with Optimal Spatio-Temporal Descriptor for General Video Recognition (CVPR 2024 · similarity 0.89)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.89)
- COSA: Concatenated Sample Pretrained Vision-Language Foundation Model (ICLR 2024 · similarity 0.89)
- Large Language Models are Good Prompt Learners for Low-Shot Image Classification (CVPR 2024 · similarity 0.89)
- LaViP: Language-Grounded Visual Prompting (AAAI 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.
