search

sanghyun-son / EDSR-PyTorch

PyTorch version of the paper 'Enhanced Deep Residual Networks for Single Image Super-Resolution' (CVPRW 2017)
MIT License
2.41k stars 666 forks source link

Why does EDSR enlarge the image twice and only low pixel images work well? #341

Open liujie316316 opened 1 year ago

liujie316316 commented 1 year ago

I use batchsize=1 to train the model, but only dozens of K images can produce results when reasoning. For example, 2M images will show insufficient memory. Why is this? image image

Zency-Sun commented 1 year ago

Have you ever run a demo to test if the program will work in your computer environment? Or is it due to lack of memory on your graphics card?

liujie316316 commented 1 year ago

Have you ever run a demo to test if the program will work in your computer environment? Or is it due to lack of memory on your graphics card?

First of all, my graphics card is 32G. It is possible to use EDSR to double the magnification of low pixels, but for example, the 1404 * 1692 image cannot be doubled. Then I cut the image into four pieces, and it is feasible to double each image. However, if you want to achieve the function of quadrupling magnification, you need to split the double image into four parts, and then enlarge each part twice to achieve quadrupling magnification. This will display the super display memory. How can I enlarge the image by four times? Is it cut into smaller pieces? Like 16 bucks?

Zency-Sun commented 1 year ago

you can add qq detailed discussion 2390641473