no code implementations • 13 Mar 2024 • Helin Cao, Sven Behnke
The images are semantically segmented by a pre-trained 2D U-Net and a dense depth prior is estimated from a depth-conditioned pipeline fueled by Depth Anything.
