Team ConcordIA - 4th place

Accuracy >96% on covidx-cxr2 test set. Score of 14.60/16 on competition test.

About our work

Go here for more details.

Requirements

• python3.8
• tqdm
• PIL
• matplotlib
• sklearn
• numpy
• torch
• torchvision

How to use

For judging:

1. Get the competition dataset from Kaggle
2. Install the requirements (see list above)
3. Run python3 eval.py --list ["<path1>", "<path2>", ...] --model <path_to_model>. The output corresponds to our final submission on eval.ai. By default, path_to_model = ./saved_models/model.pt.

Example: python3 eval.py --list ["/absolute/path/to/1.png", "/absolute/path/to/2.png", ...] --model ./saved_models/model.pt.

IMPORTANT NOTICE: note that due to the rules, you need to precisely respect the formatting (put a space between each image path, in an array, etc.).

Feel free to reach us for any question/inquiries.

Example tree:

├── against-covid-19-ConcordIA
│   ├── dataset
│   │   ├── dataset.py
│   │   ├── test.csv
│   │   └── train.csv
│   ├── learning_curves
│   │   └── <learning_curves>.png
│   ├── main.py
│   ├── models
│   │   └── resnet.py
│   ├── README.md
│   ├── saved_models
│   │   ├── sub_4.pt
│   │   └── sub_5.pt
│   ├── submissions
│   │   ├── sub_1.txt
│   │   ├── sub_2.txt
│   │   ├── sub_3.txt
│   │   ├── sub_4.txt
│   │   ├── sub_5.txt
│   │   └── sub_6.txt
│   ├── toolsp
│   │   ├── equalization.py
│   │   ├── score.py
│   │   ├── train_test.py
│   │   └── visualization.py
│   │
│   │   ### You are here
│   └── tree.txt
│
│   ### Add here the downloaded dataset
└── dataset
    ├── competition_test
    │   └── <1-400>.png
    ├── test
    │   └── <test_images>.png/jpg
    ├── test.txt
    ├── train
    │   └── <train_images>.png/jpg
    └── train.txt

For training

Download the dataset from here and extract the files to <CWD>/../dataset. Run python3 main.py to train the model, saved models are in /saved_models