Open dongzhuoyao opened 7 years ago
hi,when we use cityscapes ,the input image size of the network can not be too high,such as 400*400,however, the image size of cityscapes is 1024*2048.
so we could crop the image into many patches overlapeddly.and do inference seperately,finally do a fusion .so get the final predict result.
could you realize a interface of the process of multi-patch fusion?
hi,when we use cityscapes ,the input image size of the network can not be too high,such as 400*400,however, the image size of cityscapes is 1024*2048.
so we could crop the image into many patches overlapeddly.and do inference seperately,finally do a fusion .so get the final predict result.
could you realize a interface of the process of multi-patch fusion?