- Run the file 'StarterCode.m' to get results for 5 and 10 epochs for the two learning rates.
The files 'epoch5.mat', 'epoch10.mat', 'epoch5_learn_rate_0001.mat' and 'epoch10_learn_rate_0001.mat' contain the results of running StaterCode.m.
- 'AugmentationCode.m' contains code for generating Augmented data, which will be stored in files 'train_aug' and 'test_aug'. Run it once and save the data. The last section of this file will generate histograms of Augmentation. It uses augmentImage.m function.
The file 'augmented_data_workspace.mat' contains the results of running 'AugmentationCode.m'
-
'Aug_workspace_5_epochs.mat' and 'Aug_workspace_epoch_10.mat' contain results of running 'StarterCode.m' on the Augmented Data files.
-
'SkinnyNetwork' is the code for training the Skinny network.
The file 'skinny_workspace_attempt5.mat' contains the results of running this code. The file 'skinny_net' contains the trained model for the skinny network.
- 'WideNetwork' is the code for training the Wide network.
The file 'wide_workspace_attempt3.mat' contains the results of running this code. The file 'wide_net' contains the trained model for the wide network.
-
'SkinnyWideDataGeneration.m' code is used to generate 128 x 128 images from 256 x 256 images in the original dataset, so that they can be used on Skinny and Wide Network architectures. Run once and save the files. (The files so generated are not in separate folders for each class).
-
'skinnyTransferLearning.m' is used to train the skinny network on the original data. It uses 'skinny_net.mat' and the data generated by 'SkinnyWideDataGeneration.m'
'skinny_transferlearning_workspace.mat' has the results of the above code.
- 'wideTransferLearning.m' is used to train the wide network on the original data. It uses 'wide_net.mat' and the data generated by 'SkinnyWideDataGeneration.m'.
'wide_transferlearning_workspace.mat' has the results of the above code.
-
'ConvFilterVisualization.m' has the code to visualize all the layers of Conv and FC for all the network architectures.
-
'tsne_skinny.m' and 'tsne_wide.m' have the codes for tsne visualization of activtions of FC 2 layers of skinny and wide networks respectively.
They use 'skinny_net.mat' and 'wide_net.mat' respectively.
-
'AlexDataGeneration.m' generates 227 x 227 x 3 images from the augmented dataset for Transfer learning using Alex Net. Run this code once and save the data.
-
'AlexnetTransferLearning.m' is used for transfer learning using Alex net. It uses the data generated by 'AlexDataGeneration.m'. Alex_net_workspace.mat contains the results of running the above code.