Description
I implement light models with Pytorch, models are SqueezeNet, ShuffleNet, MobileNet, MobileNetv2 and ShuffleNetv2.
You can get details about these models at 纵览轻量化卷积神经网络：SqueezeNet、MobileNet、ShuffleNet、Xception
you can train the model with the command:
python main.py --model SuqeezeNet --epoch 100 --batch_size 64 --learning_rate 0.03 --use_cuda True
Limited by the computing power, I just verificate the correct of these models on CIFAR-10, don't get the best accuracy.
References
SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
Some parts of kuangliu's code
Some parts of togheppi's code