Open peterkim333 opened 2 years ago
Hi, how about the size of ref? You can print the size of ref to check if it matches with f_div_C.
Hi, I would like to show you. I wonder if you can run code with 512 load size.
When I modify ade20k_dataset.py for custom dataset,
parser.set_defaults(label_nc=2)
parser.set_defaults(load_size=512)
parser.set_defaults(crop_size=512)
Error: File "D:\Githubs\UNITE\models\networks\correspondence.py", line 308, in forward y1 = torch.matmul(f_divC, ref) RuntimeError: batch1 dim 2 must match batch2 dim 1
======================================================================
before operation of torch.matmul(f_divC, ref), checked sizes of each tensor like below:
f_div_C = torch.Size([1, 16384, 16384])
ref_ = torch.Size([1, 4096, 3])
When I apply size "256, 256" I can run this code without any problem The size before 'matmul' operation,
f_div_C = torch.Size([1, 4096, 4096])
ref_ = torch.Size([1, 4096, 3])
My guess is that load_size, warp_stride, and the internal fixed size of the network are intertwined. All datasets are considered to be tested only at 256 size.
Hi, I am thankful for being shared your code. I succeed in executing code with custom dataset. but, when I use large input size(from 256 to 512), I get this error
File "UNITE\models\networks\correspondence.py", line 312, in forward y1 = torch.matmul(f_divC, ref) RuntimeError: batch1 dim 2 must match batch2 dim 1
f_div_C size is doubled for width and height. if I change the tensor size, then next code makes error due to size unmatched.
I use loadsize=512 crop_size=512 label_nc = 2
please help me.
thank you.