Could you please give me some clue about how to do super-resolution task?

[juzuo]

trafficstars

Hi there, I cannot find a pretrained model to have a quick try. Could somebody give me some advice?

[rromb]

Hi there! In general, there are two straightforward approaches to the task of superresolution. Both use the same VQGAN for both the data (high resolution) and the conditioning information (low resolution):

1. Blur your input image (e.g., by downsampling and upsampling) and then concatenate the latent code of both the blurred and original images as in the other conditioning tasks.
2. Simply downsample your input image and concatenate its latent code with the latent code of the original input. This results in a shorter overall sequence and may be more efficient to train.

We will be releasing more models in the coming weeks.

[NotNANtoN]

Hi, I too would be very curious about a pretrained super-resolution network. Any plans to still release them? @rromb

[wangc-v]

Hi, i am also curious about this. @@Have you ever completed it? Could we have a comunication?@juzuo

[oguzhanbsolak]

Hi, thank you for the great work. We are also interested in the super-resolution performance of the model. Could you please, inform us if there are any released models, config files for this specific task?