Research map / Video / Video-language understanding
Image-to-Video Transfer Learning based on Image-Language Foundation Models: A Comprehensive Survey
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 |
|---|---|---|---|
| use LoRA fine-tuning instead of knowledge distillation | CLIP video-text retrieval · MSR-VTT (video-text retrieval) | R@1: 46.8 -> 50.5 | Improved |
| update CLIP parameters instead of post-network tuning on frozen features | CLIP video-text retrieval · MSR-VTT (video-text retrieval) | R@1 | Improved |
| use a stronger foundation model (Qwen2.5-VL) with simple post-network tuning | temporal video grounding · Charades-STA (TVG) | mIoU: 50.9 -> 55.9 | Improved |
| use CLIP side-tuning instead of LLaVA extra-model tuning | temporal video grounding · Charades-STA (TVG) | mIoU: 41.4 -> 50.9 | Improved |
| add video-level auxiliary models to MDETR | spatio-temporal video grounding · VidSTG declarative (STVG) | m_tIoU: 48.1 -> 52.1 | Improved |
| use LLM-based foundation model with post-network tuning instead of CLIP prompt tuning | video question answering · NExT-QA (VideoQA) | accuracy: 66.3% -> 83.73% | Improved |
| choose adaptation strategy per benchmark with fixed CLIP backbone | video question answering · MSRVTT-MC / How2QA | accuracy: 97.8 vs 71.4 | Mixed |
| disentangle spatial and temporal modeling (DiST instead of EVL) | video action recognition · Something-Something V2 (VAR) | Top-1 accuracy: 66.7 -> 73.1 | Improved |
| use adapter fine-tuning instead of side-tuning | video action recognition · Something-Something V2 (VAR) | Top-1 accuracy: 66.7 -> 70.6 | Improved |
| use knowledge distillation in open-vocabulary tracking (GLATrack) | open-vocabulary MOT · TAO (OV-MOT) | TETA: 37.9 vs 38.2 | Mixed |
Closest papers by meaning
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.89)
- PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding (NeurIPS 2025 · similarity 0.89)
- Efficient Transfer Learning for Video-language Foundation Models (CVPR 2025 · similarity 0.88)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.88)
- Knowledge Transfer from Vision Foundation Models for Efficient Training of Small Task-specific Models (ICML 2024 · similarity 0.88)
- AWT: Transferring Vision-Language Models via Augmentation, Weighting, and Transportation (NeurIPS 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.
