PeterouZh / CIPS-3D

3D-aware GANs based on NeRF (arXiv).
MIT License
610 stars 60 forks source link

Data format for training from scratch #3

Closed hoangtnm closed 2 years ago

hoangtnm commented 2 years ago

Hi, I'm quite interested in the paper. I want to know more about format of data to train the model from scratch. Do images need to be different views of the same person from different cameras, or just random 2D images with no info about yaw/pitch are enough? thank you

PeterouZh commented 2 years ago

The training images are random 2D images (FFHQ dataset) without info about yaw/pitch.