The current examples are in the form of scripts. To make easier and more interactive for users of the library it would help to have notebooks demonstrating these examples. For now the notebooks would go under examples folder under branch 2.0 where porting to Python 3+ is happening.

In augmentation, elastic_transform, it only applies a random transform on one input image array. I would think to be used for training, the image and mask pair should be transform in the same way. However, this single-input-image, single-output-image method makes it very inconvenient. Could we deform a list of images (np.arrays) using the same transformation in this method ? Thanks!

In preprocess_csv function in preprocess.py (link), pandas can be used to parse the csv more efficiently and with way less code. The machine I was using while developing the project did not have pandas installed.

It seems that you used different criteria during training and testing as the code below shows:
IN TEST:
scores = np.dot(vecs.T, qvecs)
IN TRAIN:
dif = x1 - x2
D = torch.pow(dif+eps, 2).sum(dim=0).sqrt()

   y = 0.5*lbl*torch.pow(D,2) + 0.5*(1-lbl)*torch.pow(torch.clamp(margin-D, min=0),2)
   y = torch.sum(y)

I did not get it why you do so?