---

This repository implements popular CNN architectures on CIFAR 10 dataset using pytorch.

Dependency

Pytorch (0.3.1)

Python 2.7.12

Results

CNN Architecture	Accuracy
VGG16	93.15%
VGG19	93.10%
ResNet18	93.33%
ResNet34	93.49%
ResNet101	95.11%
ResNet152	95.50%
DenseNet121	95.37%
DenseNet161	95.49%
DenseNet201	95.50%

Getting started

Train from scratch

python main.py --help
python main.py --models=densenet121 --gpu=0 --visible=0,1,2

model will be saved in checkpoints/

Fine-tuning

python main.py --models=densenet121 --gpu=0 --visible=0,1,2 --resume

Pretrained Models

CNN Architecture	Model
VGG16	vgg16
VGG19	vgg19
ResNet18	resnet18
ResNet34	resnet34
ResNet101	resnet101
ResNet152	resnet152
DenseNet121	densenet121
DenseNet161	densenet161
DenseNet201	densenet201

Author

Haibin Yu @HeroKillerEver