Research map / Video / Video-language understanding
CaReBench: A Fine-grained Benchmark for Video Captioning and Retrieval
Yifan Xu, Xinhao Li, Yichun Yang, Desen Meng, Rui Huang, Limin Wang
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 |
|---|---|---|---|
| add fine-grained alignment (stage I) to the baseline MLLM (ablation) | Qwen2-VL 7B · CaReBench captioning | Avg. F1: 26.8 -> 33.8 | Improved |
| add fine-grained alignment (stage I) to the baseline MLLM (ablation) | Qwen2-VL 7B · CaReBench retrieval | Avg. R@1: 25.6 -> 17.6 | Hurt |
| add retrieval adaptation (stage II contrastive NLI training) (ablation) | Qwen2-VL 7B · CaReBench retrieval | Avg. R@1: 25.6 -> 77.0 | Improved |
| add retrieval adaptation (stage II) to the baseline (ablation) | Qwen2-VL 7B · CaReBench captioning | Avg. F1: 26.8 -> 28.2 | Improved |
| add fine-grained alignment on top of retrieval adaptation (ablation) | CaRe · CaReBench retrieval | Avg. R@1: 77.0 -> 78.0 | Improved |
| add retrieval adaptation (stage II) after stage I (ablation) | CaRe stage-I · CaReBench captioning | overall action F1: 35.3 -> 35.1 | No clear effect |
| add contrastive training on NLI data to MLLMs | Qwen2-VL 7B · CaReBench General Retrieval text-to-video | R@1: 30.9 -> 76.6 | Improved |
| use spatial rather than temporal captions as queries | all models · CaReBench spatial vs temporal retrieval | R@1: CaRe 76.8 -> 50.7 text-to-video (temporal lower) | Improved |
Closest papers by meaning
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.89)
- VERIFIED: A Video Corpus Moment Retrieval Benchmark for Fine-Grained Video Understanding (NeurIPS 2024 · similarity 0.89)
- Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers (CVPR 2024 · similarity 0.89)
- Needle In A Video Haystack: A Scalable Synthetic Evaluator for Video MLLMs (ICLR 2025 · similarity 0.89)
- AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmark (ICLR 2025 · similarity 0.88)
- SF2T: Self-supervised Fragment Finetuning of Video-LLMs for Fine-Grained Understanding (CVPR 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.
