search

joeyballentine / 2C2-ESRGAN

ESRGAN (Enhanced Super Resolution GAN) using two 2x2 (kernel size) conv2d layers instead of a traditional single 3x3 conv2d layer in its conv block.
MIT License
13 stars 1 forks source link

About model training #2

Open shamelesslyAI opened 1 year ago

shamelesslyAI commented 1 year ago

Hello! I was wondering if you have information about the training sources of the model? I'm looking for upscalers trained with cc0/public domain images only, but can't tell if this one qualifies for the purpose.

Thank you!

joeyballentine commented 1 year ago

The pretrained model I released for this was trained with the Nomos2k dataset, which is a combination of images from Adobe-MIT-5k, RAISE, FFHQ, DIV2K, DIV8k, Flickr2k, Rawsamples, SignatureEdits, Hasselblad raw samples and Unsplash (according to the author of the dataset).

Hope that helps.