PAMI: Part Anchored Motion for Text to Human-Object Interaction Generation
PAMI generates text-conditioned human-object interactions by voting object motion from body-part anchors.
PAMI generates text-conditioned full-body human-object interaction by localizing object motion through votes from multiple body-part anchors, inspired by the Hough transform. PamiVAE learns an interaction latent space and decodes frame-wise weights that combine those votes. PamiGen first produces a coarse interaction from text, and PamiRefiner then resolves contact with hybrid long-range probes and short-range surface sensors. On InterAct, the method reports more accurate human-relative object motion and 14.5% higher contact recall than the previous state of the art.
- Object motion is expressed relative to multiple body-part anchors.
- PamiVAE learns frame-wise weights that aggregate part-specific votes.
- PamiGen drafts coarse motion; PamiRefiner fixes fine contact geometry.
- On InterAct, contact recall is 14.5% above the prior state of the art.
From the original article
1 min read at huggingface.co
Text-conditioned full-body human-object interaction (HOI) generation requires synthesizing human motion and object trajectories that match the input text while remaining precisely coordinated over time. Most methods represent the human and object as separate trajectories and predict the global human-object couplings.
By Chuqiao Li, Xianghui Xie, Yong Cao · 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.