README

Mask Image Classification

Code for solution in Mask Image Classification hosted by Naver Boostcamp AI Tech.

To learn more detail about the competition, please, refer to the AI Stage post

Archive contents

minibatch28/
├── data/
|   ├── image/
|   |   ├── train/ 
│   |   |   ├── 00001_male_Asian_40/
│   |   |   |   ├── mask.jpg
│   |   |   |   ├── mask2.jpg
│   |   |   |   ├── incorrect.png 
│   |   |   |   └── normal.jpeg 
│   |   |   ├── {Number}_{Gender}_{Race}_{Age}/
│   |   |   |   ├── ...
│   |   |   |   └── ...
|   |   |   └── 99999_female_Asian_150/
│   |   |       ├── mask.jpg 
│   |   |       ├── incorrect.jpg
│   |   |       └── normal.jpg 
|   |   └── eval/ 
│   |       ├── abcde.jpg
│   |       ├── {Any_image_name}.jpg
│   |       └── lorem_ipsum.jpeg 
│   ├── train.csv
│   └── info.csv
├── output/
│   └── ensemble/ 
├── models/
├── dataset.py
├── loss.py
├── inference.py
├── train.py
└── train.sh

data/ : contains raw data dir and label data (should contain 'train.csv', 'info.csv')
data/image/ : raw image dir of the competition
data/eval/ : evaluation image dir of the competition
output/ : inference result csv files will be created
output/ensemble/ : ensemble result csv files will be created
models/ : contains trained state_dict of each model

Requirements

Ubuntu 18.04.5 LTS
Python 3.8.5
Pytorch 1.7.1
CUDA 11.0

You can use the pip install -r requirements.txt to install the necessary packages.

Hardware

CPU: 8 x Intel(R) Xeon(R) Gold 5120 CPU @ 2.20GHz
GPU: 1 x Tesla V-100
RAM: 88G

Prepare Data

You can automatically generate data/train_list.csv, data/valid_list.csv files by running train.py

Train Model

To train model, run following command.

$ python train.py --model {model_number} --dataset {model_number} --batch_size {batch_size} --epochs {epochs} \

    --lr_decay_step {lr_decay_step} --gamma {gamma} --lr {learning_rate} --scheduler 1 \

    --cutmix 0 --criterion {model_number} --optimizer {optimizer}

To train 5 models at once, run following shell script file.

$ ./train.sh

Predict

If trained weights are prepared, you can create files that contains class of images.

$ python inference.py --model {model_number} --batch_size {batch_size}

Ensemble

To inference and ensemble 5 models at once, run following shell script file.

$ ./ensemble.sh

Then ensemble.csv will be created in output/ensemble directory.

Members of Minibatch 28

강재현_T2003
권예환_T2012
김준태_T2058
박마루찬_T2078
서광채_T2106
장동건_T2185
홍요한_T2244

yh00214 / image-classification-level1-28