Skip to content

Event

Holi-World: Building World Intelligence from First-Person Experience

1 reports · 1 sources

The authors of the Holi-World paper propose the HOLI-WORLD integrated ecosystem, which connects human first-person experience collection, world representation, automated data construction, model training, and independent evaluation. HOLIWORLD-Foundry combines static and dynamic perception to automatically construct world records; HOLIWORLD-Data, based on approximately 100 self-collected scenes, compiles about 80K QA pairs for training HOLIWORLD-VL-4B; and independently human-annotated held-out recordings constitute HOLIWORLD-Bench. The paper states that experiments on ScanNet, ScanNet++, Aria Digital Twin, and HOLIWORLD-Bench show that the system performs strongly in static perception and dynamic tracking.

Report timeline

  1. Holi-World: Building Holistic World Intelligence from Human First-Person Experience

    The study proposes the HOLI-WORLD integrated ecosystem, which connects experience collection, world representation, automated data construction, model learning, and independent evaluation. HOLIWORLD-Foundry combines static and dynamic perception to automatically construct world records, and HOLIWORLD-Data organizes approximately 100 self-collected scenes into approximately 80K QA pairs used to train HOLIWORLD-VL-4B; independently human-annotated held-out recordings form HOLIWORLD-Bench.

    HuggingFace Daily Papers ·

Related topics