transfLearn

Based on: http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/

1. Download the weight at http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/bvlc_alexnet.npy (they need to be in the working directory)
2. Download the DRIVE database: http://www.isi.uu.nl/Research/Databases/DRIVE/

Structure

myalexnet_forward.py contains the pretrained network.

input_reader.py is the module to retrieve inputs and labels

svm.py contains the svm classifier

Usage

See svm.py as example

Use test = input_reader.create_dataset() to extract NUM_TRIAL random batches from the DRIVE dataset of size BATCH_[HEIGHT|WITDH] resized to ALEX_NET_[HEIGHT|WIDTH]

Use test = input_reader.prepare_image(filename,labelname) to split an image in to n batches of size BATCH_[HEIGHT|WITDH] resized to ALEX_NET_[HEIGHT|WIDTH]

Use features, labels = net.extract_features(test) to extract the 'fc7' layer of AlexNet and the corresponding label

Use train(features, labels) to train the svm

Use prediction = predict(features) to classify a new instance