Pytorch Implementation of Deep Learning Models/Algorithms

This repositiory is for implementing and training/testing popular model architectures on the CIFAR10 dataset.

Environment

• CUDA Version: 10.2

torch==1.5.0
torchvision==0.6.0
numpy==1.19.2

Usage

Training

To train a model, run train.py. If you need to speicfy the model, just use some args.

# train alexnet model with using gpu. 50 epochs
$ python train.py --model alexnet --epoch 50 --gpu

optional&required arguments

--data_dir      default='./data/train',
                help="Directory containing the dataset"
--model         required=True, type=str,
                help="The model you want to train"
--lr            type=float, default=0.001,
                help="Learning rate"
--epoch         type=int, default=50,
                help="Total training epochs"
--batch_size    type=int, default=256,
                help="batch size"
--gpu           action='store_true', default='False',
                help="GPU available"

Evaluate

To evaluate the model, run evaluate.py. If you need to speicfy the model, just use some args.

# evaluate alexnet model
$ python evaluate.py --model alexnet --weights ./results/alexnet/best.pth --gpu

optional&required arguments

--data_dir      default='./data/test',
                help="Directory containing the dataset"
--model         required=True, type=str,
                help="The model you want to test"
--weight        required=True,
                help="The weights file you want to test"
--batch_size    default=256,
                help="batch size"
--gpu           action='store_true', default='False',
                help="GPU available"

Results

Network	epoch	lr	top1@prec(test)	ModelSize(MB)
AlexNet	50	0.001	74.2578%	266MB
ZFNet	50	0.01	80.4395%	445MB
VGG	-	-	-	-
ResNet	-	-	-	-
Inception	-	-	-	-
GoogLeNet	-	-	-	-
-	-	-	-	-
-	-	-	-	-
-	-	-	-	-
-	-	-	-	-