VinAIResearch / CREPS

Efficient Scale-Invariant Generator with Column-Row Entangled Pixel Synthesis (CVPR 2023)
https://vinairesearch.github.io/CREPS/
GNU Affero General Public License v3.0
29 stars 1 forks source link

How to generate high resolution images #1

Closed jerrymatjila closed 1 year ago

jerrymatjila commented 1 year ago

Thanks for the paper and the code. Kindly share coding guidelines of using a CREPS model trained on 512×512 images to generate near-realistic images at higher resolution. Lets say 1024x1024

thuanz123 commented 1 year ago

Hi @jerrymatjila, sorry for my late reply. With the latest update, you now can generate images at any size (larger than 0 of course). For example:

python gen_images.py --outdir=out --trunc=0.7 --seeds=600 --network=creps-ffhq-512x512.pkl --scale=2

This command will produce an image at resolution 1024x1024

thuanz123 commented 1 year ago

Apart from scaling, you can manipulate the variable "G.synthesis.b4.input.coords" to get your desired geometric transformation like translation, rotation, or distortion

jerrymatjila commented 1 year ago

Thanks