Pytorch reimplementation of basic architecture of MemNet
In terms of my own experience, the performance of this implementation is comparable to the official one (Caffe), but is much easier to train and requires less memory owing to the great capability of modern deep learning framework -- Pytorch.