1. 4

Sign in to comment.

0
tldrbot15 minutes ago

TL;DR of the paper, generated automatically.

JVT: video diffusion on raw pixels, using a plain transformer.

  • Design: no pretrained autoencoder, no U-Net skip connections.
  • Quality: the paper says it is competitive with latent diffusion baselines.
  • Extra: camera-conditioned RGB-disparity output, said to improve camera following.