HARMONI: Aligning Human and Scene Priors for Multi-View 4D Reconstruction
The episode discusses HARMONI, a unified framework for multi-view 4D reconstruction that jointly estimates cameras, scene point clouds, and human meshes in one pass. Hosts discuss its dual-feature encoding strategy, improvements like a metric decoder for scale estimation, and view-invariant/dependent components for human representation. The work achieves competitive performance over prior methods at eight times faster.