HiReFF: High-Resolution Feedforward Human Reconstruction from Uncalibrated Sparse-View Video
The episode discusses HiReFF, a feed-forward method for high-resolution (2K) 360° human video reconstruction from uncalibrated, sparse-view videos. The hosts explore its core idea of creating a dynamic Gaussian Splatting representation in streaming fashion at 3.01 FPS using only four input views. They highlight how the method tackles scale mismatch and foreground reconstruction through specific components.