"index_select_out_cuda_impl" not implemented for 'Float'

[seogihyun]

Hi, I tried to execute demo version.

python3 main.py --demo --demo_dir='dataset' --resume='model_best.pth' --STN --mask

But there is one issue.

So, first I found the same issue in this github like this.

But when I try to execute like that, there was another issue. I think the problem is torch version with cudnn. So please let me know how to solve this problem.

Thank you.

[E-qin]

maybe you can try to change the data type of that tensor.

[Everythingismetaphor]

I find this problem too.

[LuThanhThien]

maybe you can try to change the data type of that tensor.

you can use this method from @E-qin to ensure predecessors is type of 'torch.long()' or simply correct the devision operator (/) into floor devision operator (//):

predecessors = (candidates / selfl.num_classes .... -> predecessors = (candidates // selfl.num_classes ....