hamer101/fungus

Added Transfoermer, DS-MIL

fungus

Database

Database can be acquired via: https://drive.google.com/file/d/1-CjXE-HO3xz36x_XNJVDTRV89xDnjLlu/view?usp=sharing

Conda environment

Conda environment can be set up by issuing:

conda env create -f environment.yml
conda develop .

Training

Firstly, you need to calculate parameters for normalization. You can do that by invoking

python dataset/normalization.py PATH_TO_IMAGES

Results will be saved in tmp/{means.npy,stds.npy}.

Then issue

python experiments/extract_features.py PATH_TO_IMAGES PATH_TO_MASKS --prefix PREFIX --size PATCH_SIZE --model MODEL
python experiments/extract_features.py PATH_TO_IMAGES PATH_TO_MASKS --prefix PREFIX --size PATCH_SIZE --model MODEL --test
python experiments/hyperparameters.py --prefix PREFIX --features FEATURES --model MODEL
python experiments/confusion_matrices.py --prefix PREFIX --features FEATURES --model MODEL

Parameters meaning can be checked in each script's documentation.

Evaluation

In order to inspect our model check out showcase/examine_model.ipynb.

Other

To run the experiments with neural networks, use branch feature/neural_networks.

To obtain the analysis of SVM classifier use notebook SVM analysis.ipynb.