This is pytorch implemention of mobile architecture，converted from gluon model_zoo

Note：

• The architecture is coming from paper, so the original image size is 224x224 and with rescale=32. however, here use cifar10 as training dataset (image size is 32x32, and resize to 64x64), so the accuracy in this data is not well. (just as demo, so I did not try to modify the architecture)
• train code is just demo，not compare the architecture
• The default dataset root in your_computer_name/data

change model name：

# choose network --- choose 0
model_name = ['mobilenet_v1', 'mobilenet_v2', 'shufflenet_v1', 'shufflenet_v2'][0]

accuracy after epoch80：89.48%

change model name：

# choose network --- choose 1
model_name = ['mobilenet_v1', 'mobilenet_v2', 'shufflenet_v1', 'shufflenet_v2'][1]

accuracy after epoch80：89.29%

change model name：

# choose network --- choose 2
model_name = ['mobilenet_v1', 'mobilenet_v2', 'shufflenet_v1', 'shufflenet_v2'][2]

accuracy after epoch80：85.44%

change model name：

# choose network --- choose 3
model_name = ['mobilenet_v1', 'mobilenet_v2', 'shufflenet_v1', 'shufflenet_v2'][3]

accuracy after epoch80：82.99%

1. gluon model_zoo：mobilenetv1&v2
2. pytorch-cifar：shufflenet（nearly all the code is copy）