Research map / Video / Video-language understanding
Fine-Tuned CLIP Models Are Efficient Video Learners
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Fine-tune both CLIP encoders on video with temporal embedding pooling | CLIP ViT-B/16 · HMDB-51 zero-shot | accuracy: 44.6 -> 51.3 | Helped |
| Fine-tune both CLIP encoders on video with temporal embedding pooling | CLIP ViT-B/16 · UCF-101 zero-shot | accuracy: 72.0 -> 76.8 | Helped |
| Fine-tune both CLIP encoders on video with temporal embedding pooling | CLIP ViT-B/16 · Kinetics-400 base-to-novel | Harmonic mean: 64.0 -> 67.9 | Helped |
| Fine-tune both CLIP encoders on video with temporal embedding pooling | CLIP ViT-B/16 · HMDB-51 few-shot, 2-shot | accuracy: 53.0 -> 57.2 | Helped |
| Fine-tune both CLIP encoders on video with temporal embedding pooling | CLIP ViT-B/16 · Kinetics-400 fully supervised | Top-1 accuracy: 84.7 -> 83.9 | Hurt |
| Add embedding-level fusion rather than decision-level fusion (ablation) | ViFi-CLIP · Kinetics-400 tiny | accuracy: 74.6 -> 76.9 | Helped |
| Add embedding-level fusion rather than image-level fusion (ablation) | ViFi-CLIP · SSv2 few-shot, 16-shot | accuracy: 11.2 -> 12.4 | Helped |
| Add embedding-level fusion rather than decision-level fusion (ablation) | ViFi-CLIP · HMDB-51 few-shot, 16-shot | accuracy: 62.2 -> 66.8 | Helped |
| Add embedding-level fusion rather than image-level fusion (ablation) | ViFi-CLIP · UCF-101 few-shot, 16-shot | accuracy: 91.7 -> 92.7 | Helped |
| Use bridge-and-prompt adaptation with vision and language prompts after video fine-tuning | CLIP · HMDB-51 few-shot, 2-shot | accuracy: 60.5 -> 63.0 | Helped |
| Use bridge-and-prompt adaptation with vision and language prompts after video fine-tuning | CLIP · UCF-101 few-shot, 2-shot | accuracy: 89.0 -> 91.0 | Helped |
Closest papers by meaning
- ViLT-CLIP: Video and Language Tuning CLIP with Multimodal Prompt Learning and Scenario-Guided Optimization (AAAI 2024 · similarity 0.90)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.88)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.88)
- Boosting the visual interpretability of CLIP via adversarial fine-tuning (ICLR 2025 · similarity 0.88)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 2025 · similarity 0.88)
- Demystifying CLIP Data (ICLR 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.