solangii / CIFAR10-CIFAR100

image classification for CIFAR-10, CIFAR-100 using pytorch

Course project of CSE463 Machine Learning, UNIST

Image Classification for CIFAR-10, CIFAR-100

Platform

This code was implemented and tested with below environment

python == 3.6.12
torch == 1.9.0
torchvision == 0.10.0

Training

# ************************** CIFAR-10 *****************************
$ python train.py --model lenet --dataset cifar10 --n_train 20000 
$ python train.py --model MLP --dataset cifar10 --n_train 20000 
$ python train.py --model resnet --dataset cifar10 --n_train 20000 
$ python train.py --model vgg --dataset cifar10 --n_train 20000 
$ python train.py --model googlenet --dataset cifar10 --n_train 20000 
# ************************** CIFAR-100 *****************************
$ python train.py --model lenet --dataset cifar100 --n_train 20000 
$ python train.py --model MLP --dataset cifar100 --n_train 20000 
$ python train.py --model resnet --dataset cifar100 --n_train 20000 
$ python train.py --model vgg --dataset cifar100 --n_train 20000 
$ python train.py --model googlenet --dataset cifar100 --n_train 20000 

Performance

• lr : 0.005, epochs: 100, batch_size : 100, optimizer : SGD, n_train : 20000

Dataset	Model	Accuracy
CIFAR-10	RF	42.19 %
CIFAR-10	MLP	44.3 %
CIFAR-10	LeNet	64 %
CIFAR-100	RF	16.7 %
CIFAR-100	MLP	2.45 %
CIFAR-100	LeNet	24 %

• learning rate
  • epochs: 100, batch_size : 100, optimizer : SGD, n_train : 20000

Dataset	Learning Rate	Accuracy
CIFAR-10	0.1	25 %
CIFAR-10	0.01	61 %
CIFAR-10	0.005	64 %
CIFAR-10	0.001	55 %

• CNN

  • lr : 0.005, epochs: 100, batch_size : 100, optimizer : SGD, n_train : 20000

  Dataset	Model	Accuracy
  CIFAR-10	LeNet-5	64 %
  CIFAR-10	GoogLeNet	74 %
  CIFAR-10	VGG19	78 %
  CIFAR-10	ResNet18	74 %

• Dataset

  • lr : 0.005, epochs: 100, batch_size : 100, optimizer : SGD, n_train : 20000

  Dataset	Model	Accuracy
  CIFAR-10	LeNet-5	64 %
  CIFAR-100	LeNet-5	24 %
  Fasion MNIST	LeNet-5	85 %

• number of training data
  • lr : 0.005, epochs: 100, batch_size : 100, optimizer : SGD

Dataset	Model	# of data	Accuracy
CIFAR-10	LeNet-5	10000	56 %
CIFAR-10	LeNet-5	20000	64 %
CIFAR-10	LeNet-5	30000	66 %
CIFAR-10	LeNet-5	40000	66 %
CIFAR-10	LeNet-5	50000	69 %