Research map / Video / Video-language understanding
IF-VidCap: Can Video Caption Models Follow Instructions?
Shihao Li, Yuanxing Zhang, Jiangtao Wu, Zhide Lei, Yiwen He, Runzhe Wen, Chenxi Liao, Chengkang Jiang, An Ping, Shuo Gao, Suhan Wang, Zhaozhou Bian and 11 more
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 |
|---|---|---|---|
| fine-tune Qwen2.5-VL-7B-Instruct on the 46K instruction-following caption triplets | Qwen2.5-VL-7B-Instruct · IF-VidCap | ISR / CSR: ISR 10.92 -> 12.76; CSR 58.12 -> 61.64 | Improved |
| scale Qwen2.5-VL from 3B to 72B | Qwen2.5-VL-Instruct · IF-VidCap | ISR: 6.54 -> 17.50 | Improved |
| enable thinking mode | InternVL-3.5 8B · IF-VidCap | ISR: 9.96 -> 17.33 | Improved |
| enable thinking mode | VideoLLaMA3 7B · IF-VidCap | ISR: 10.63 -> 12.21 | Improved |
| use dense-captioning specialist (Tarsier2) instead of general MLLM | video captioning model · IF-VidCap | ISR: Tarsier2 1.40 vs Qwen2.5-VL-7B 10.92 | Hurt |
| increase frames from 8 to 64 then to 128 (ablation) | Qwen2.5-VL-7B-Instruct · IF-VidCap | ISR / CSR | Mixed |
| increase input resolution from 168x168 to 784x784 (ablation) | Qwen2.5-VL-7B-Instruct · IF-VidCap | ISR / CSR | Improved |
| increase instruction complexity (more constraints) (ablation) | evaluated MLLMs · IF-VidCap curated 1021 subset | CSR / ISR | Hurt |
| use GPT-5-mini as judge instead of open-source judges (ablation) | IF-VidCap evaluator · IF-VidCap | agreement with human: 92.03 / 92.18 -> 96.33 | Improved |
Closest papers by meaning
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.90)
- Video ReCap: Recursive Captioning of Hour-Long Videos (CVPR 2024 · similarity 0.88)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.88)
- ViDiC: Video Difference Captioning (NeurIPS 2026 · similarity 0.88)
- AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmark (ICLR 2025 · similarity 0.88)
- VidChain: Chain-of-Tasks with Metric-based Direct Preference Optimization for Dense Video Captioning (AAAI 2025 · 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.
