Hi.
I have a question about image size in this track. Are the given HR images provided before cropping?
I just found that for example, 0007x8 is (255x169) shape. So in that case, I think HR image of 0007 has to be (2040x1352) but it is (2040x1356).
If this right, in my opinion, it could be a problem in the case when evaluate using some portion of training dataset to tune hyper parameter.
It will be very helpful if you check me whether my thought is wrong or not.
Thanks.
Posted by: nmhkahn @ Jan. 24, 2018, 7:16 p.m.I have the same question.
Posted by: zzsmg @ Feb. 4, 2018, 1:32 a.m.Hi everyone， thank you for the question.
Before we downscale the original image to generate the training image,
we first crop the boundary pixels,
y = y(1:height- mod(height, scale), 1:width - mod(width, scale), :);
I hope this can help.