Arguments: download_data # data_dir # reconfigure_val model # pretrained # name init #

This program can be run on python 2 or 3.

pip install -r requirements.txt

python train.py --download_data

This will download the data from web, and start running my implementation, ExtResNet-50, with all the default parameters.

Once the data is downloaded in data/tiny-imagenet-200/tiny-imagenet-200/, every subsequent execution can be performed using the command: python train.py

usage: train.py [-h] [--download_data] [--data_dir DATA_DIR] [--reconfigure_val] [--model MODEL] [--pretrained] [--name NAME] [--init INIT] [--activation ACTIVATION] [--loss LOSS] [--do DO] [--lr LR] [--kernel_size KERNEL_SIZE] [--kernel_number KERNEL_NUMBER] [--data_aug DATA_AUG]

optional arguments: -h, --help show this help message and exit --download_data Set if app needs to download tiny imagenet dataset from web. --data_dir DATA_DIR Directory in which the input data is stored. --reconfigure_val Set if val dataset should be reconfigured. --model MODEL Name of the architecture to train and validate the data. One of [vgg16/resnet50/ext-resnet50 --pretrained Use imagenet pretrained model if set. Not applicable if model = ext-resnet50. --name NAME Name of this training run. Will store results in {output/log/weights}/[model]/[name]. --init INIT Kernel initializer of the hidden layers. [random- uniform/glorot-uniform/he-uniform]. Used only when [model] is [vgg16/ext-resnet50]. --activation ACTIVATION The activation function of the network. One of [relu/softsign/tanh]. Used only when [model] is [ext- resnet50]. --loss LOSS The loss function of the network. One of [categorical_ crossentropy/mean_squared_error/categorical_hinge]. Used only when [model] is [ext-resnet50]. --do DO The dropout at the output of each block. Used only when [model] is [ext-resnet50]. --lr LR The learning rate of the network. Used only when [model] is [ext-resnet50]. --kernel_size KERNEL_SIZE The kernel size for the network. One of [same/halved]. Used only when [model] is [ext-resnet50]. --kernel_number KERNEL_NUMBER The number of kernels at each layer of the network. One of [same/doubled]. Used only when [model] is [ext- resnet50]. --data_aug DATA_AUG Data augmentation to perform. One of [basic/no/yes].

python train.py --model ext-resnet62

python train.py --model vgg16 --pretrained

To predict on a test data set, type the following command: python predict.py

usage: predict.py [-h] [--model MODEL] [--data_dir DATA_DIR] [--weights WEIGHTS] [--activation ACTIVATION] [--do DO] [--kernel_size KERNEL_SIZE] [--kernel_number KERNEL_NUMBER] [--data_aug DATA_AUG]

optional arguments: -h, --help show this help message and exit --model MODEL Name of the architecture to train and validate the data. One of [vgg16/resnet50/ext-resnet50/ext-resnet41 /ext-resnet62]. --data_dir DATA_DIR Directory in which the test data is stored. --weights WEIGHTS Path of the weights file to load into the model. --activation ACTIVATION The activation function of the network. One of [relu/softsign/tanh]. Used only when [model] is [ext- resnet50]. --do DO The dropout at the output of each block. Used only when [model] is [ext-resnet50]. --kernel_size KERNEL_SIZE The kernel size for the network. One of [same/halved]. Used only when [model] is [ext-resnet50]. --kernel_number KERNEL_NUMBER The number of kernels at each layer of the network. One of [same/doubled]. Used only when [model] is [ext- resnet50]. --data_aug DATA_AUG Data augmentation to perform. One of [basic/no/yes].

To plot confusion matrix on a validation data set, type the following command: python conf_matrix.py

usage: conf_matrix.py [-h] [--num_classes NUM_CLASSES] [--model MODEL] [--data_dir DATA_DIR] [--weights WEIGHTS] [--name NAME] [--activation ACTIVATION] [--do DO] [--kernel_size KERNEL_SIZE] [--kernel_number KERNEL_NUMBER]

optional arguments: -h, --help show this help message and exit --num_classes NUM_CLASSES Number of classes --model MODEL Name of the architecture to train and validate the data. One of [vgg16/resnet50/ext-resnet50/ext-resnet41 /ext-resnet62]. --data_dir DATA_DIR Directory in which the val data is stored. --weights WEIGHTS Path of the weights file to load into the model. --name NAME Name of the confusion matrix and image file --activation ACTIVATION The activation function of the network. One of [relu/softsign/tanh]. Used only when [model] is [ext- resnet50]. --do DO The dropout at the output of each block. Used only when [model] is [ext-resnet50]. --kernel_size KERNEL_SIZE The kernel size for the network. One of [same/halved]. Used only when [model] is [ext-resnet50]. --kernel_number KERNEL_NUMBER The number of kernels at each layer of the network. One of [same/doubled]. Used only when [model] is [ext- resnet50].