nothinglo/Deep-Photo-Enhancer

arbitrary resolutions input

xuanzhangyang opened this issue · 2 comments

paper said: Although the model was trained on 512x512 inputs, we have extended it so that it can handle arbitrary resolutions1.
how to extend it ?

The simple implementation is resizing the input image for the global feature but using the full-size for the local feature and then combine them to the U-Net decoder.

The simple implementation is resizing the input image for the global feature but using the full-size for the local feature and then combine them to the U-Net decoder.

Hello, could you please give us more info how to achieve this? Our images are all 8K ... Many thanks!!