AI RESEARCH
GenFusion: Feed-forward Human Performance Capture via Progressive Canonical Space Updates
arXiv CS.CV
•
ArXi:2603.28997v1 Announce Type: new We present a feed-forward human performance capture method that renders novel views of a performer from a monocular RGB stream. A key challenge in this setting is the lack of sufficient observations, especially for unseen regions. Assuming the subject moves continuously over time, we take advantage of the fact that body parts become observable by maintaining a canonical space that is progressively updated with each incoming frame.