trqminh / fashion-MNIST

experiments on fashion mnist

fashion-MNIST

computer vision course assignment

version 1

Model:

input
  |
  v
conv1 (5x5x32)
  |
  v
max_pooling (2x2)
  |
  v
conv2 (5x5x64)
  |
  v
max_pooling (2x2)
  |
  v
conv3 (5x5x128)
  |
  v
max_pooling (2x2)
  |
  v
fc1 (128*2*2, 120)
  |
  v
fc2 (120, 84)
  |
  v
fc3 (84, 10)
  |
  v
Cross Entropy

optimizer things

• learning rate: 0.001
• activation: ReLU
• Dropout, BatchNorm
• Augment: random horizontal flip

Accuracy on test set

93.76996805111821 %

version 2

Model

Ensemble 10 model in version 1 by incremental their output value

optimizer things

• learning rate: 0.001
• activation: ReLU
• Dropout, BatchNorm
• Augment: random horizontal flip

Accuracy on test set

94.27915335463258 %

version 3

Model

Wide-ResNet

input
  |
  v
conv1
  |
  v
resnet block1
  |
  v
resnet block2
  |
  v
resnet block3
  |
  v
 fc
  |
  v
Cross Entropy

optimizer things

• learning rate: 0.001
• activation: ReLU
• Dropout, BatchNorm
• Augment: random horizontal flip

Accuracy on test set

94.68849840255591 %

Usage

Requirements

• Python 3
• CUDA (optional)

Download repository

git clone https://github.com/trqminh/fashion-MNIST.git
cd fashion-MNIST

Install libraries:

pip3 install -r requirements.txt

Install pytorch-cpu:

pip3 install torch==1.2.0+cpu torchvision==0.4.0+cpu -f https://download.pytorch.org/whl/torch_stable.html

or install pytorch with cuda (if you have already installed CUDA)

pip3 install torch==1.2.0+cu92 torchvision==0.4.0+cu92 -f https://download.pytorch.org/whl/torch_stable.html

create data/ directory in repository 's root directory, put the csv files in it (download from here)

create trained_models/ directory in repository 's root directory and put the .pth files in it (download from here)

Directory structure

|-- data
|   |-- fashion-mnist_test.csv
|   |-- fashion-mnist_train.csv
|-- models
|   |-- __init__.py
|   |-- my_model.py
|   |-- wide_resnet.py
|-- trained_models
|   |-- version1_model.pth
|   |-- version2_model.pth
|   |-- version3_model.pth
|-- utils
|   |-- __init__.py
|   |-- custom_data.py
|-- .gitignore
|-- README.md
|-- requirements.txt
|-- test.py
|-- train.py

Training

• Train the model in each version, with number of epochs (Recommend install CUDA)

python3 train.py --version $version --epoch $epoch

• Example:

python3 train.py --version 3 --epoch 10

Evaluate

• Test my trained models in each version

python3 test.py --version $version

• Example:

python3 test.py --version 3
Evaluating...
Accuracy of the network on 10000 test images: 94.68849840255591 %