hongdd/cifar10.classifer

Use pytorch to implement some classifer network

CNN Classifer In Cifar-10

I try to do some classifer experiment in cifar-10. All the network train with the same parameter,training from scratch. I want to know in the same condition which CNN is the best.

TODO

• VGG
• ResNet
• MobileNet
• ShuffleNet
• MobileNetV2
• ResNeXt
• InceptionV3
• ShuffleNetV2
• SENet

Experiment:

Training Detail

• training from scratch
• epoch 100
• init lr 0.01
• cosine lr scheduler
• single GTX 1080

Model	train time(min)	ACC
VGG16	74m 26s	91.55
ResNet50	XXXXXXXXXXXXXX	X
MobileNet	XXXXXXXXXXXXXX	X
MobileNetV2	XXXXXXXXXXXXXX	X
ShuffleNet	XXXXXXXXXXXXXX	X
ResNeXt50_32x4d	XXXXXXXXXXXXXX	X

Reference:

https://github.com/kuangliu/pytorch-cifar