Research map / Video / Video-language understanding
ScaleLong: A Multi-Timescale Benchmark for Long Video Understanding
David Ma, Huaqing Yuan, Xingjian Wang, Qianbo Zang, Tianci Liu, Xinyang He, Yanbin Wei, Jiawei Guo, nijiahui, Zhenzhu Yang, Meng Cao, Shanghaoran Quan and 7 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 |
|---|---|---|---|
| increase input frames from 4 to 128 at 240p (ablation) | MLLMs · ScaleLong Clip | accuracy (%): 43.5 -> 54.5 | Improved |
| increase input frames from 4 to 128 at 240p (ablation) | MLLMs · ScaleLong Story | accuracy (%): 42.5 -> 49.5 | Improved |
| increase frames from 64 to 128 (ablation) | MLLMs · ScaleLong Event | accuracy (%): peaks at 64 frames, slightly declines at 128 | Hurt |
| raise resolution from 72p to 360p at 32 frames (ablation) | MLLMs · ScaleLong Clip | accuracy (%): roughly 2% absolute gain | Improved |
| raise resolution from 360p to 480p at 32 frames (ablation) | MLLMs · ScaleLong Clip | accuracy (%): slight dip | Hurt |
| raise frames instead of resolution (ablation) | MLLMs · ScaleLong Clip | accuracy (%): about 9% jump vs 2% from resolution | Improved |
| use 32x240p allocation at fixed token budget (ablation) | Qwen2.5-VL · ScaleLong Clip | accuracy (%): peaks at 53.6 | Improved |
| use 8x480p instead of 4x720p-type low-frame allocation (ablation) | Qwen2.5-VL · ScaleLong Shot | accuracy (%): 40.2 -> 45.5 | Improved |
| scale InternVL2.5 from 8B to 78B (ablation) | InternVL2.5 · ScaleLong Shot | accuracy (%): 42.5 to 54.3 | Improved |
| scale InternVL2.5 from 8B to 78B (ablation) | InternVL2.5 · ScaleLong Story | accuracy (%): 52.3 to 61.5 | Improved |
| use Gemini 2.5 Pro instead of best open-source model | Gemini 2.5 Pro · ScaleLong Event | accuracy (%): 14.6 points higher than InternVL2.5-78B | Improved |
Closest papers by meaning
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.93)
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.92)
- TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding (CVPR 2024 · similarity 0.92)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.92)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.92)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.91)
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.
