Holi-World builds world records and a 4B model from ego video
Original title: Holi-World: Building Holistic World Intelligence from Human-Ego Experience
About 100 scenes yield roughly 80K QA pairs used to train HOLIWORLD-VL-4B, evaluated on HOLIWORLD-Bench.
The paper introduces HOLI-WORLD, linking capture, world representation, automated data construction, learning, and evaluation. HOLIWORLD-Foundry builds records connecting scene structure, object states, and interaction histories. HOLIWORLD-Data uses about 100 self-collected scenes and about 80K QA pairs to train HOLIWORLD-VL-4B. Training improved spatial and temporal reasoning on HOLIWORLD-Bench and EgoDynamic4D.
- Foundry combines static and dynamic perception to build records automatically.
- Held-out human annotations form HOLIWORLD-Bench.
- Static perception and tracking were tested on ScanNet, ScanNet++, and Aria Digital Twin.
- Training improved reasoning on HOLIWORLD-Bench and EgoDynamic4D.
From the original article
1 min read at huggingface.co
Human egocentric experience offers a rich foundation for learning how the world is organized and how it changes through interaction. Realizing this potential at scale requires turning growing volumes of recordings into learning resources that connect scene structure, object semantics, and interaction histories across viewpoints and time, while reducing dependence on costly manual annotation.
By Hao Li, Xiyuan Zheng, Yuanyuan Gao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.