SRGAN Custom dataset preparation

Hello, I am wondering what points we should care of, in custom dataset preparation. Whats in my mind is like:

Resolution of HR image set. Are there any restrictions in size; or things to change in the code for larger images?
Image aspect ratio
Recommended minimum number of images Or any other points... Thank you.

Mar 09 '18 18:03 ontheway16

The demo code use image with width and heigh around 1200~2000, and crop sub-image with 384x384.
No limit.
1000 images should be good.

Mar 12 '18 14:03 wagamamaz

@wagamamaz Thank you for the info. For 1st question, is this means we cannot use anything larger then 2000px? If this is the case, whats needed to alter this to something larger? (or restricted by GPU memory maybe?) Sorry I am not into this (yet), so my questions may sound primitive for the moment.

Mar 12 '18 14:03 ontheway16

@ontheway16 For 1st question, you can use larger image but don't be too large, because but the current model is designed for the demo dataset.

If you want larger image, 1) change the model 2) crop larger sub-image 3) larger GPU memory

Mar 12 '18 15:03 wagamamaz

Thank you very much, I will investigate.

Mar 12 '18 17:03 ontheway16

What is the logic behind cropping the image by 384x384 ? I want to train the network with 3840x2160 px images, what should be the ideal size of my cropped sub-image?

Mar 22 '18 21:03 rahat-yasir

i think 384x384 should be fine for 3840x2160

Mar 22 '18 21:03 zsdonghao

@zsdonghao Thank you for sharing the very concise implementation. I am working on a dataset where the image scale factor from HR/LR is 1.5 and 2.25.
I am having a hard wrapping my head around how would I get it to work on these data. I am limited as the original HR images are in 4:3 aspect ratio and need to be upsampled in the same aspect ratio ..

Any thoughts on how this would come across in the SRGAN model ?

Thanks.

Sep 29 '18 00:09 amanmeetgarg