Pytorch_CNN_Architectures

My PyTorch implementation of CNN models. All models are trained with the CIFAR-100 dataset.

Credits

Some of the codes in this repository are written by weiaicunzai.

Also, please cite me if you use the codes in this repository.

Contribution

Please feel free to contribute to this repository.

Requirements

Below are the specifications of my experiment environment:

• python3.6
• pytorch@1.0
• tensorflow@1.5(optional)
• cuda@8.0
• cudnnv5
• tensorboardX@1.6(optional)

Perhaps, pytorch@>=0.4 would also be fine (I am not 100% sure).

Usage

Dataset

I will use cifar100 dataset from torchvision since it's more convenient, but I also kept the sample code for writing your own dataset module in dataset folder, as an example for people don't know how to write it.

Running the tensorbard(optional)

Install tensorboardX (a tensorboard wrapper for pytorch)

$ pip install tensorboardX
$ mkdir runs
Run tensorboard
$ tensorboard --logdir='runs' --port=6006 --host='localhost'

Training the model

Train all the model on a Tesla P40(22912MB)

You need to specify the net you want to train using arg -net

$ python train.py -net vgg16

sometimes, you might want to use warmup training by set -warm to 1 or 2, to prevent network diverge during early training phase.

The supported net args are:

• squeezenet
• mobilenet
• mobilenetv2
• shufflenet
• shufflenetv2
• vgg11
• vgg13
• vgg16
• vgg19
• densenet121
• densenet161
• densenet201
• googlenet
• inceptionv3
• inceptionv4
• inceptionresnetv2
• xception
• resnet18
• resnet34
• resnet50
• resnet101
• resnet152
• preactresnet18
• preactresnet34
• preactresnet50
• preactresnet101
• preactresnet152
• resnext50
• resnext101
• resnext152
• attention56
• attention92
• seresnet18
• seresnet34
• seresnet50
• seresnet101
• seresnet152
• nasnet

Normally, the weights file with the best accuracy would be written to the disk with name suffix 'best'(default in checkpoint folder).

Testing the model

Test the model using test.py file. By replacing the network name and file path of the weights file, you could test other networks with your own weights.

$ python test.py -net vgg16 -weights path_to_vgg16_weights_file

Implementated NetWork

• vgg Very Deep Convolutional Networks for Large-Scale Image Recognition
• googlenet Going Deeper with Convolutions
• inceptionv3 Rethinking the Inception Architecture for Computer Vision
• inceptionv4, inception_resnet_v2 Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
• xception Xception: Deep Learning with Depthwise Separable Convolutions
• resnet Deep Residual Learning for Image Recognition
• resnext Aggregated Residual Transformations for Deep Neural Networks
• resnet in resnet Resnet in Resnet: Generalizing Residual Architectures
• densenet Densely Connected Convolutional Networks
• shufflenet ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
• shufflenetv2 ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design
• mobilenet MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• mobilenetv2 MobileNetV2: Inverted Residuals and Linear Bottlenecks
• residual attention network Residual Attention Network for Image Classification
• senet Squeeze-and-Excitation Networks
• squeezenet SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
• nasnet Learning Transferable Architectures for Scalable Image Recognition

Training Details

I didn't use any training tricks to improve accuray, if you want to learn more about training tricks, please refer to this repository, which contains various common training tricks and their pytorch implementations.

Basically, I followed the hyperparameter settings in paper Improved Regularization of Convolutional Neural Networks with Cutout, which is init lr = 0.1 divide by 5 at 60th, 120th, 160th epochs, train for 200 epochs with batchsize 128 and weight decay 5e-4, Nesterov momentum of 0.9. You could also use the hyperparameters from paper Regularizing Neural Networks by Penalizing Confident Output Distributions and Random Erasing Data Augmentation, which is initial lr = 0.1, lr divied by 10 at 150th and 225th epochs, and training for 300 epochs with batchsize 128, this is more commonly used. You could decrese the batchsize to 64 or whatever suits you, if you dont have enough gpu memory.

You can choose whether to use TensorBoard to visualize your training procedure.

Results

The result I can get from a certain model, since I use the same hyperparameters to train all the networks, some networks might not get the best result from these hyperparameters, you could try yourself by finetuning the hyperparameters to get better result.

dataset	network	params	top1 err	top5 err	memory	epoch(lr = 0.1)	epoch(lr = 0.02)	epoch(lr = 0.004)	epoch(lr = 0.0008)	total epoch
cifar100	mobilenet	3.3M	34.02	10.56	0.69GB	60	60	40	40	200
cifar100	mobilenetv2	2.36M	31.92	09.02	0.84GB	60	60	40	40	200
cifar100	squeezenet	0.78M	30.59	8.36	0.73GB	60	60	40	40	200
cifar100	shufflenet	1.0M	29.94	8.35	0.84GB	60	60	40	40	200
cifar100	shufflenetv2	1.3M	30.49	8.49	0.78GB	60	60	40	40	200
cifar100	vgg11_bn	28.5M	31.36	11.85	1.98GB	60	60	40	40	200
cifar100	vgg13_bn	28.7M	28.00	9.71	1.98GB	60	60	40	40	200
cifar100	vgg16_bn	34.0M	27.07	8.84	2.03GB	60	60	40	40	200
cifar100	vgg19_bn	39.0M	27.77	8.84	2.08GB	60	60	40	40	200
cifar100	resnet18	11.2M	24.39	6.95	3.02GB	60	60	40	40	200
cifar100	resnet34	21.3M	23.24	6.63	3.22GB	60	60	40	40	200
cifar100	resnet50	23.7M	22.61	6.04	3.40GB	60	60	40	40	200
cifar100	resnet101	42.7M	22.22	5.61	3.72GB	60	60	40	40	200
cifar100	resnet152	58.3M	22.31	5.81	4.36GB	60	60	40	40	200
cifar100	preactresnet18	11.3M	27.08	8.53	3.09GB	60	60	40	40	200
cifar100	preactresnet34	21.5M	24.79	7.68	3.23GB	60	60	40	40	200
cifar100	preactresnet50	23.9M	25.73	8.15	3.42GB	60	60	40	40	200
cifar100	preactresnet101	42.9M	24.84	7.83	3.81GB	60	60	40	40	200
cifar100	preactresnet152	58.6M	22.71	6.62	4.20GB	60	60	40	40	200
cifar100	resnext50	14.8M	22.23	6.00	1.91GB	60	60	40	40	200
cifar100	resnext101	25.3M	22.22	5.99	2.63GB	60	60	40	40	200
cifar100	resnext152	33.3M	22.40	5.58	3.18GB	60	60	40	40	200
cifar100	attention59	55.7M	33.75	12.90	3.47GB	60	60	40	40	200
cifar100	attention92	102.5M	36.52	11.47	3.88GB	60	60	40	40	200
cifar100	densenet121	7.0M	22.99	6.45	1.28GB	60	60	40	40	200
cifar100	densenet161	26M	21.56	6.04	2.10GB	60	60	60	40	200
cifar100	densenet201	18M	21.46	5.9	2.10GB	60	60	40	40	200
cifar100	googlenet	6.2M	21.97	5.94	2.05GB	60	60	40	40	200
cifar100	inceptionv3	22.3M	22.81	6.39	2.26GB	60	60	40	40	200
cifar100	inceptionv4	41.3M	24.14	6.90	4.11GB	60	60	40	40	200
cifar100	inceptionresnetv2	65.4M	27.51	9.11	4.14GB	60	60	40	40	200
cifar100	xception	21.0M	25.07	7.32	1.67GB	60	60	40	40	200
cifar100	seresnet18	11.4M	23.56	6.68	3.12GB	60	60	40	40	200
cifar100	seresnet34	21.6M	22.07	6.12	3.29GB	60	60	40	40	200
cifar100	seresnet50	26.5M	21.42	5.58	3.70GB	60	60	40	40	200
cifar100	seresnet101	47.7M	20.98	5.41	4.39GB	60	60	40	40	200
cifar100	seresnet152	66.2M	20.66	5.19	5.95GB	60	60	40	40	200
cifar100	nasnet	5.2M	22.71	5.91	3.69GB	60	60	40	40	200