We have made public 10% of our data. For full licensing, please contact Yi Zhou at yizho@adobe.com .

This work is done during my internship at @adobe.com with Yi Zhou @papaginayi.bsky.social , Chunhao Huang, Uttaran Bhattacharya, and Qixing Huang.

HUMOTO also enables researchers to generate views from any angle, with controlled lighting, scenes, and backgrounds - perfect for training models that understand realistic human-object relationships 👁 and creating authentic visual content 🎨 with proper physical interactions!

HUMOTO captures functional, purposeful grasps showing how humans naturally manipulate objects in daily tasks - perfect for teaching robots 🤖 human-like manipulation skills and developing more natural virtual character animations 💃🏻.

🎬 We build the dataset like directing a movie! Scene-driven LLM scripting created natural purposeful tasks. Mocap suits + dual cameras help capture detailed motions. 🎯 Meticulous cleaning ensures minimal penetration and jittering with quality control metrics.