Closed yingqichao closed 3 years ago
The most likely reason is insufficiently high resolution. The method is designed for spatially large images only, because patches would start overlapping otherwise. Check the data loader code for more details regarding size. Does switching dataset help?
Thanks for your reply. I find the minimum required image size is 768*768, otherwise the patch-wise detector cannot work.
Just what kind of images are qualified as "high-quality"?
Error:
File "/home/qichaoying/Downloads/dissecting-image-crops-main/data.py", line 555, in getitem raise RuntimeError('Could not load a suitable input image after 32 tries.') RuntimeError: Could not load a suitable input image after 32 tries.
I have tried MS-COCO(jpg) and CelebA(png) as training set.