AndrewYB / AlexNet-Experiments

Experiments with AlexNet architecture for Image Classification

AlexNet Experiments

This repository contains a PyTorch implementation of the AlexNet architeture described in ImageNet Classification with Deep Convolutional Neural Networks. The following experiments have been performed on the architecture:

• Activation Unit: ReLU vs Tanh
• Dropout v/s No Dropout
• Overlapping Pooling v/s Non-Overlapping Pooling
• Optimization Techniques: SGD v/s SGD with momentum v/s SGD with momentum and weight decay v/s ADAM.

The experiments were done on a 35 class subset of the ImageNet Dataset that can be downloaded from here.

Organisation

The code in the repository are organised as follows:

• network.py: AlexNet arch. implementation
• dataset.py: custom Pytorch Dataset implementation
• train.py: training/validation/testing code
• main.py: driver code

Requirements

• python 3.5
• pytorch

The dataset may be downloaded from here.

Usage

Use the following to get help with usage.

python main.py --help

Results

The results of all the experiments can are compiled in results.pdf. The best model that attained over 76% accuracy on the the test-set can be downloaded from here.