Hi. I loaded the pwc_net.pth.tar with the pytorch model and ran the inference on all the MPI Single dataset. I used the following way to calculate the average EPE loss:
for each image:epe = torch.norm(flow_gt-flo, p=2, dim=0).mean()epe_list.append(epe)
(after the loop:)
epe_list.mean()
where dim=0 is the axis corresponding to flow channels and has size 2.
And I get a score of 3.20 on Clean and 3.61 on Final . This is different from the numbers you reported for MPI Sintel. Could you please help me realize what I might be doing wrong? Thank you.
Hi. I loaded the
pwc_net.pth.tarwith the pytorch model and ran the inference on all the MPI Single dataset. I used the following way to calculate the average EPE loss:for each image:epe = torch.norm(flow_gt-flo, p=2, dim=0).mean()epe_list.append(epe)(after the loop:)
epe_list.mean()where dim=0 is the axis corresponding to flow channels and has size 2. And I get a score of 3.20 on Clean and 3.61 on Final . This is different from the numbers you reported for MPI Sintel. Could you please help me realize what I might be doing wrong? Thank you.