SimpleNet Classifier

This repo would allow you to train an image classifier with Tensorflow Keras. The model architecture is based of SimpleNet. https://arxiv.org/abs/1608.06037

Setup

How to train

Go to config to adjust your training parameters. Choice of MNIST or CIFAR10 dataset for training, and choice of data augmentation. Use help function to view arg options. To train custom dataset, include 'data' folder with directory structure:

data/
	train/
		class_a/
			class_a01.jpg
			class_a02.jpg
			...
		class_b/
			class_b01.jpg
			class_b02.jpg
			...
	test/
		class_a/
			class_a01.jpg
			class_a02.jpg
			...
		class_b/
			class_b01.jpg
			class_b02.jpg

Arguments

-a, --data_aug: whether to initialize data augmentation
-s, --dataset: choose 'mnist' or 'cifar10'\

Command

python training.py

Evaluation

Load saved model by specifying model path in argument. Also allows loading of custom dataset and data augmentation. Classification report and confusion matrix is saved in results folder, and models are saved in saved_models folder

Arguments

-w, --model_path: directory to model file (.hdf5)
-d, --data_path: directory to data folder
-a, --data_aug: where to initialize data augmentation

Command

python evaluation.py