Research map / Video / Video-language understanding
Token Mixing: Parameter-Efficient Transfer Learning from Image-Language to Video-Language
Yuqi Liu, Luhui Xu, Pengfei Xiong, Qin Jin
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 token mixing to a transformer layer (ablation) | partially fine-tuned BLIP ViT-B/16 · MSVD | CIDEr: 146.3 -> 146.9 | Improved |
| add token mixing to a transformer layer (ablation) | partially fine-tuned BLIP ViT-B/16 · MSRVTT | R@1: 46.1 -> 47.1 | Improved |
| freeze visual projection instead of fine-tuning it (ablation) | Token Mixing on BLIP ViT-B/16 · MSVD | CIDEr: 146.92 -> 138.87 | Hurt |
| freeze language head instead of fine-tuning it (ablation) | Token Mixing on BLIP ViT-B/16 · MSVD | CIDEr: 146.92 -> 141.15 | Hurt |
| freeze both attention and FFN in the token-mixing layer (ablation) | Token Mixing on BLIP ViT-B/16 · MSVD | CIDEr: 146.92 -> 134.39 | Hurt |
| freeze FFN in the token-mixing layer (ablation) | Token Mixing on BLIP ViT-B/16 · MSVD | CIDEr: 146.92 -> 144.45 | Hurt |
| freeze attention in the token-mixing layer (ablation) | Token Mixing on BLIP ViT-B/16 · MSVD | CIDEr: 146.92 -> 145.46 | Hurt |
| apply token mixing rather than spatial-only attention (ablation) | fully fine-tuned BLIP ViT-B/16 · MSVD | CIDEr: 150.1 -> 151.2 | Improved |
| apply token mixing rather than spatial-only attention (ablation) | fully fine-tuned BLIP ViT-B/16 · MSRVTT | R@1: 47.6 -> 48.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SwinBERT — SwinBERT: End-to-End Transformers With Sparse Attention for Video Captioning (CVPR 2022) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization (ICML 2024 · similarity 0.90)
- READ-PVLA: Recurrent Adapter with Partial Video-Language Alignment for Parameter-Efficient Transfer Learning in Low-Resource Video-Language Modeling (AAAI 2024 · similarity 0.87)
- Language Model Beats Diffusion - Tokenizer is key to visual generation (ICLR 2024 · similarity 0.87)
- Efficient Transfer Learning for Video-language Foundation Models (CVPR 2025 · similarity 0.87)
- Video Token Merging for Long Video Understanding (NeurIPS 2024 · similarity 0.87)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · 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.
