Training FFHQR dataset using FFHQ dataset as input

[luser350]

Hi, thanks for sharing your great work. I want to train a stylegan2-ada to take input from an FFHQ image and generate an FFHQR image. The FFHQR dataset is a retouched version of FFHQ.

To my understanding, I can train a FFHQR model using the following command:

python train.py --outdir=training-runs --data=ffhqr/  --gpus=1 --resume=ffhq1024 --aug=noaug

However, this model will learn to generate FFHQR images from noise. The different images can be produced by using a different seed with generate.py

So how can I use stylegan2-ada-pytorch for such img2img translation problem? Your few words can help me a lot. Thanks!!!