About the logging the warped image about sfmlearner-pytorch

[ClementPinard]

hello, the :1 here refers to the fact that for logging purpose only, we construct a fake batch of only one element, to make things faster. You could argue that it could have been done directly from the actual batch computing in the photometric loss function. It's actually what I did in this version of the code : https://github.com/ClementPinard/unsupervised-depthnet/blob/master/train_img_pairs.py#L381

I did not put it here because it needs a bit of refactoring (but it might come !)

as such everything behaves like the photometric loss function here : https://github.com/ClementPinard/SfmLearner-Pytorch/blob/master/loss_functions.py#L26

ref_imgs is a list of imgs tensors (from t-2 to t+2, excluding t), pose is a tensor of poses, hence the necessity of the enumerate in the line before.

But here, ref is the jth element in the list of ref_imgs, and thus it matches with the element pose[:,j]

[gengrui1983]

Thank you @ClementPinard !