Research map / Video / Video-language understanding
Kangaroo: A Powerful Video-Language Model Supporting Long-context Video Input
Jiajun Liu, Yibing Wang, Hanghang Ma, Xiaoping Wu, Xiaoqi Ma, Xiaoming Wei, Jianbin Jiao, Enhua Wu, Jie Hu
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 |
|---|---|---|---|
| Kangaroo (8B) versus VideoLLaMA2 8x7B on long video | VideoLLaMA2 · MLVU | accuracy: 35.5 -> 61.0 | Improved |
| Kangaroo versus the best prior 7B/8B model Video-CCAM on MLVU | Video-CCAM · MLVU | accuracy: 58.5 -> 61.0 | Improved |
| Kangaroo versus Gemini-1.5-Pro on LVBench | Gemini-1.5-Pro · LVBench | accuracy: 33.1 -> 39.4 | Hurt |
| Kangaroo versus GPT-4V on EgoSchema | GPT-4V · EgoSchema | accuracy: 55.6 -> 62.7 | Improved |
| Kangaroo versus GPT-4V on MVBench | GPT-4V · MVBench | accuracy: 43.7 -> 61.1 | Improved |
| Kangaroo versus Video-CCAM on MVBench | Video-CCAM · MVBench | accuracy: 64.60 -> 61.1 | Hurt |
| Kangaroo (64 frames) versus LongVA (128 frames) on VideoMME | LongVA · VideoMME (overall w/o subs) | accuracy: 52.6 -> 56.0 | Improved |
| Kangaroo versus VITA 8x7B on VideoMME with subtitles | VITA · VideoMME (overall w subs) | accuracy: 57.6 -> 57.6 | No clear effect |
| Kangaroo versus GPT-4V on SeedBench video | GPT-4V · SeedBench video | accuracy: 61.7 -> 59.2 | Hurt |
| Kangaroo versus LLaVA-NeXT-IL on SeedBench video | LLaVA-NeXT-IL · SeedBench video | accuracy: 53.7 -> 59.2 | Improved |
Closest papers by meaning
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.90)
- Understanding Long Videos with Multimodal Language Models (ICLR 2025 · similarity 0.89)
- MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding (CVPR 2024 · similarity 0.89)
- World Model on Million-Length Video And Language With Blockwise RingAttention (ICLR 2025 · similarity 0.89)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.89)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.89)
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.
