CNN Classifier

This repo includes easy-to-read scripts to train, validate and evaluate your own dataset using any of the most popular architectures, including: AlexNet, VGG, SqueezeNet, ResNet, and DenseNet, in any of their configurations (eg. DenseNet201, ResNet152, VGG16_BN, etc).

The whole code is written in:

Requirements:

• Package requirements are the same for pytorch.
• Works either on python 2 or 3.
• The whole code will assume you have your dataset in ./data with subfolders: train, val, and test. If you need to change this, just modify data_loader.py.

Usage:

./main.py --kwargs

kwargs:

• batch_size [default=128]
• num_epochs [default=59]
• num_epochs_decay [default=60]
• stop_training [default=3]
• num_workers [default=4]
• model [default='densenet201']
• TEST [default=False]

Example:

./main.py --batch_size=16 --model=resnet152

Misc

It trains using pretrained weights from Imagenet.