AlexNet Implementation with Pytorch

• Unofficial implementation of the paper ImageNet Classification with Deep Convolutional Neural Networks

0. Develop Environment

Docker Image
- tensorflow/tensorflow:tensorflow:2.4.0-gpu-jupyter

Library
- Pytorch : Stable (1.7.1) - Linux - Python - CUDA (11.0)

• Using Single GPU

1. Implementation Details

• model.py : AlexNet model
• train.py : train AlexNet (include 10-crop on val/test)
• utils.py : count correct prediction
• AlexNet - Cifar 10.ipynb : install library, download dataset, preprocessing, train and result
• Visualize - Kernel.ipynb : visualize the first conv layer
• Details
  • Follow ImaegNet train details : batch size 128, learning rate 0.01, momentum 0.9, weight decay 0.0005
  • No learning rate scheduler for convenience
  • No augmentation using PCA
  • Different network initialization strategy as paper
  • Different image pre-processing as paper (use CIFAR 10 statistics)

2. Result Comparison on CIFAR-10

Source	Score	Detail
Paper	87	without normalization
Paper	89	with normalization
Current Repo	89.47	with normalization

3. Reference

• ImageNet Classification with Deep Convolutional Neural Networks [paper]