search

Valentyn1997 / xray

Unsupervised Anomaly Detection for X-Ray Images
Other
45 stars 15 forks source link

Use Project #2

Closed nabinabila closed 3 years ago

nabinabila commented 3 years ago

I wanna use the project, how could i run it and have models results (the visualization and evaluation metrics) ?

Valentyn1997 commented 3 years ago

I wanna use the project, how could i run it and have models results (the visualization and evaluation metrics) ?

In the most basic setting, you would need to:

  1. Install all the libraries, listed in requirements.txt.
  2. Download MURA dataset and put it to data/train folder, in the project root directory.
  3. Start MlFlow server.
  4. Run train.py (https://github.com/Valentyn1997/xray/blob/master/src/models/train.py) and monitor models on MlFlow server.
  5. Use saved models for visualisations (see notebooks https://github.com/Valentyn1997/xray/tree/master/notebooks/models) and evaluation (see notebook https://github.com/Valentyn1997/xray/blob/master/notebooks/Experiments_comparison.ipynb)

There are also some scripts for data preprocessing in https://github.com/Valentyn1997/xray/tree/master/src/features (e.g. cropping and hand-detection).

viettr commented 3 years ago

Regarding point 4: You need to specify the model in line 27. Currently you can choose between: AlphaGan, BottleneckAutoencoder, BaselineAutoencoder, DCGAN, SAGAN and VAE.

Valentyn1997 commented 3 years ago

Do you mean, you need specific model weights? Unfortunately, we don't anymore store the weights of the models, reported in the paper.

viettr commented 3 years ago

I mean one needs to specify the model one wants to train in the line 27 in the train.py file.