Cifar100-Classification

My play ground with image classification for Cifar100 dataset with various models

Usage

• first you need to get the package

$ git clone https://github.com/dolong2110/Cifar100-Classification.git

• Then make sure you are in right directory

$ cd Cifar100-Classification

• Install all the requirement packages

$ pip install -r requirements.txt

• finally train the model

$ python3 train.py --model resnet18 --image_size 32 --augmentation True

Notes here is that:

• resnet18 is the model's name you can replace its with any available models in my package. For instance resnet152, mobilenet, etc.
• 32 is the images' size. It is the default image size of the cifar100.
• True here is whether we should augment data or not.

Models Using

• self-implement cnn (basic_nn)
• linear regression (linear_regression)
• resnet9 (resnet9)
• resnet18 (resnet18)
• resnet34 (resnet34)
• resnet50 (resnet50)
• resnet101 (resnet101)
• resnet152 (resnet152)
• mobilenet (mobilenet)
• mobilenetv2 (mobilenetv2)

Report

1. Version 1

Model	Accuracy
basic_nn
linear_regression
resnet9	0.6188
resnet18	0.6405
resnet34	0.6479
resnet50	0.6133
mobilenet
mobilenetv2	0.4572

2. Version2

add data augmentation

Model	Accuracy
basic_nn
linear_regression
resnet9	0.6375
resnet18	0.6739
resnet34	0.6909
resnet50
mobilenet	0.4635
mobilenetv2

3. Version3

add epoch from 10 to 20

Model	Accuracy
basic_nn
linear_regression
resnet9
resnet18
resnet34
resnet50
mobilenet
mobilenetv2