Deep Learning with PyTorch

Extracting convolutional features

We pass the data from the train and validation data loaders through the model and store the results of the model in a list for further computation. By calculating the pre-convoluted features, we can save a lot of time in training the model, as we will not be calculating these features in every iteration. In the following code, we calculate the pre-convulted features:

#For training data# Stores the labels of the train datatrn_labels = [] # Stores the pre convoluted features of the train datatrn_features = [] #Iterate through the train data and store the calculated features and the labelsfor d,la in train_loader:    o = m(Variable(d.cuda()))    o = o.view(o.size(0),-1)    trn_labels.extend(la) trn_features.extend(o.cpu().data) ...