Question about training dataset

Open nahyunKim2221 opened this issue 5 years ago • 1 comments

Hi, I have a question about generating patches at training.

In your paper, you use CBSD500 as training dataset and make patches which have 64x64 shape. I wonder the way of generating patches.

Do you use overlapping? or any other tricks? And, How many patches per a image is used in training?

Thanks.

Dec 04 '20 07:12 nahyunKim2221

We use the sliding window to crop the image patches(64*64) from the original BSD500 dataset with the overlapping of 32. In total, about 63000 image patches were generated. You can also use any other overlapping, like 48, 32, 16 .etc, to control the number of image patches you use in training.

Best regards.

Dec 04 '20 08:12 zhwhong