Research map / Video / Video-language understanding
Language Models with Image Descriptors are Strong Few-Shot Video-Language 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 |
|---|---|---|---|
| add object tokens to frame captions (ablation) | VidIL · MSVD-QA val | accuracy (%): 39.6 -> 40.3 | Helped |
| add event tokens on top of frame+object (ablation) | VidIL · MSVD-QA val | accuracy (%): 40.3 -> 39.9 | No clear effect |
| add attribute tokens on top of frame+object (ablation) | VidIL · MSVD-QA val | accuracy (%): 40.3 -> 40.9 | Helped |
| use four frames instead of one frame (ablation) | VidIL · MSVD-QA val | accuracy (%): 38.5 -> 40.8 | Helped |
| keep correct temporal order instead of reversing (ablation) | VidIL · MSVD-QA val | accuracy (%): 40.7 -> 40.8 | No clear effect |
| dynamic in-context example selection at 5 examples (ablation) | VidIL · MSVD-QA val | accuracy (%): 38.4 -> 40.4 | Helped |
| select 5 examples out of 30 shots vs using all 30 (ablation) | VidIL · MSVD-QA val | accuracy (%): 40.0 -> 41.1 | Helped |
| select 5 examples out of 20 shots vs using all 20 (ablation) | VidIL · MSVD-QA val | accuracy (%): 42.6 -> 42.2 | No clear effect |
| add ASR transcripts to prompt | VidIL · YouCook2 caption | CIDEr: 27.0 -> 111.6 | Helped |
| train retrieval model on VidIL pseudo labels instead of other captioners' pseudo labels | BLIP · MSR-VTT retrieval | video-to-text R@1: 33.2 -> 39.6 | Helped |
| train on UniVL pseudo labels | BLIP · VaTeX retrieval | video-to-text R@1: 28.2 -> 25.5 | Hurt |
| few-shot image-language prompting vs supervised MERLOT | VidIL · VLEP | accuracy (%): 68.4 -> 72.0 | Helped |
Closest papers by meaning
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.92)
- OmniViD: A Generative Framework for Universal Video Understanding (CVPR 2024 · similarity 0.89)
- Mimir: Improving Video Diffusion Models for Precise Text Understanding (CVPR 2025 · similarity 0.89)
- ViLMA: A Zero-Shot Benchmark for Linguistic and Temporal Grounding in Video-Language Models (ICLR 2024 · similarity 0.89)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.88)
- LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment (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.