Deep Learning with PyTorch

Extracting DenseNet features

It is quite similar to what we did for Inception, except we are not using register_forward_hook to extract features. The following code shows how the DenseNet features are extracted:

#For training datatrn_labels = []trn_features = []#code to store densenet features for train dataset.for d,la in train_loader:    o = my_densenet(Variable(d.cuda()))    o = o.view(o.size(0),-1)    trn_labels.extend(la)    trn_features.extend(o.cpu().data)#For validation dataval_labels = []val_features = []#Code to store densenet features for validation dataset. for d,la in val_loader:    o = my_densenet(Variable(d.cuda()))    o = o.view(o.size(0),-1)    val_labels.extend(la)    val_features.extend(o.cpu().data)

The preceding code is similar to what we have ...