AI Paper Atlas

Research map / Video / Video generation

MotionAtlas: A High-Quality Dataset and Benchmark for Dense Motion Captioning

Weisong Liu, Haochen Wang, Gaokuan Gaokuan, Yuhao Wang, Yikang Zhou, Zhongwei Ren, Guangcan Mai, Anran Wang, Yanwei Li, Xiangtai Li, Zhaoxiang Zhang

ECCV 2026 · Video › Video generation · approach: Compositional text-to-video diffusion · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
fine-tune on MotionAtlas-DataQwen3-VL-4B · MotionAtlas-Bench single-frameaccuracy: 19.3 -> 27.7Improved
fine-tune on MotionAtlas-DataQwen3-VL-8B · MotionAtlas-Bench single-frameaccuracy: 24.3 -> 31.6Improved
fine-tune on MotionAtlas-DataMolmo2-4B · MotionAtlas-Bench single-frameaccuracy: 13.7 -> 21.6Improved
fine-tune on region-level MotionAtlas-DataQwen3-VL-4B · TOMATOaccuracy: 27.4 -> 35.2Improved
fine-tune on MotionAtlas-DataQwen3-VL-4B · FAVOR-Benchaccuracy: 47.0 -> 55.0Improved
fine-tune on MotionAtlas-DataQwen3-VL-8B · NExT-QAaccuracy: 76.9 -> 77.2No clear effect
add region detail captions with visual cue instead of general captions only (ablation)Qwen3-VL-4B SFT · TOMATOaccuracy: 28.4 -> 35.2Improved
add region detail captions (text reference) instead of general captions only (ablation)Qwen3-VL-4B SFT · FAVOR-Benchaccuracy: 52.2 -> 55.7Improved
use visual cue instead of text reference for region (ablation)Qwen3-VL-4B SFT · FAVOR-Benchaccuracy: 55.7 -> 55.0Mixed
add self-bootstrap refinement (ablation)MA data pipeline · MotionAtlas-Benchaccuracy: 36.4 -> 39.9Improved
add full-video caption (ablation)MA data pipeline · MotionAtlas-Benchaccuracy: 33.2 -> 39.9Improved
add spatial crop (ablation)MA data pipeline · MotionAtlas-Benchaccuracy: 32.7 -> 39.9Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy