tensorlayer / SRGAN

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
https://github.com/tensorlayer/tensorlayerx
3.24k stars 813 forks source link

How can build a custom dataset? #250

Open huchi00057 opened 1 year ago

huchi00057 commented 1 year ago

In my enormous tests, I found it's neccesory to resize image to be multiple of 3. (600300 or 999123)

I wanna make sure whether there is any principle that i need to follow?

Like image format should be .jpg ? Or justify batch size ? and so on...

I truely cannot execute train.py succesfully. Orz😡😡😡

It just showed that the memory is not enough, and it cannot allocate approximately 17000 bytes in my custom dataset. But it ran smoothly in the original data provided in Github.

huchi00057 commented 1 year ago

PS. I use RTX 2070i to execute it, and image size of train set is 1249*1146 pixels while its amount is 800. I don't think there is any some wrong.πŸ™„πŸ™„