/cnn-architectures

use keras to implement the main convolutional neural networks

CNN Architectures

Datasets

  • CIFAR10: default dataset
  • Animals: 3 classes: cats, dogs and pandas

Hyperparameter Configuration

  • optimization method: SGD
  • learning rate: 0.01
  • weight decay: None/0.0005
  • batch size: 32/64/128
  • epochs: 10/30

Ordered by Timeline

  • 2012.09 AlexNet CoLab
  • 2013.11 ZFNet CoLab
  • 2014.09 VGGNet CoLab
  • 2014.09 GoogeLeNet CoLab
  • 2015.12 Inception V2/V3 CoLab
  • 2015.12 ResNet V1
  • 2016.02 Inception-v4 Colab
  • 2016.02 SqueezeNet Colab
  • 2016.03 ResNet V2 Colab
  • 2016.10 Xception Colab
  • 2016.11 ResNeXt Colab
  • 2017.04 MobileNet Colab
  • 2017.09 SENet
  • 2017.06 ShuffleNet Colab
  • 2018.01 MobileNet V2 Colab
  • 2018.06 ShuffleNet V2 Colab