Pytorch implementation for few-shot photorealistic video-to-video translation. It can be used for generating human motions from poses, synthesizing people talking from edge maps, or turning semantic label maps into photo-realistic videos. The core of video-to-video translation is image-to-image translation. Some of our work in that space can be found in pix2pixHD and SPADE.
Few-shot Video-to-Video Synthesis
Ting-Chun Wang, Ming-Yu Liu, Andrew Tao, Guilin Liu, Jan Kautz, Bryan Catanzaro
NVIDIA Corporation
In Neural Information Processing Systems (NeurIPS) 2019
- Dance Videos
- Talking Head Videos
- Street View Videos
