But looking at the results, the two values were different.
line 425, to_optimise2 = [0.2001, 0.4251, 0.0000]
line 439, to_optimise = [0.2001, 0.4251, 0.4891]
I don't understand what the code wrote implicitly.
combined = torch.cat((identity_reprojection_loss, reprojection_loss), dim=1)
https://github.com/nianticlabs/monodepth2/issues/264 In other issues, auto mask is mentioned to summarize the code below.
min_reprojection_loss, idx1 = torch.min(reprojection_loss, dim=1) min_identity_reprojection_loss, idx2 = torch.min(identity_reprojection_loss, dim=1) automask = (min_reprojection_loss < min_identity_reprojection_loss).float() to_optimise2 = min_reprojection_loss * automask
But looking at the results, the two values were different. line 425, to_optimise2 = [0.2001, 0.4251, 0.0000] line 439, to_optimise = [0.2001, 0.4251, 0.4891]
I don't understand what the code wrote implicitly.
combined = torch.cat((identity_reprojection_loss, reprojection_loss), dim=1)