Oulu-IMEDS / AdaTriplet

AdaTriplet

Dependencies

We provide file environment.yml for dependencies

Installation

  conda env create -f environment.yml
  conda activate AdaTriplet
  cd AdaTriplet/
  pip install -e .

Experiments

Dataset

1. Knee X-ray: Download the OAI dataset via https://nda.nih.gov/oai/.
2. Chest X-ray: Download the ChestXrays-14 dataset at https://nihcc.app.box.com/v/ChestXray-NIHCC/
3. Create the image folder ./ResizedImages_{OAI|CXR}

Training

Run the script as follows:

  python train.py data_type=<OAI|CXR> \ 
        image_crop_path=<path-to-ResizedImages> \  
        image_raw_path=<path-to-folder-of-raw-images> \
        metadatapath=<path-to-meta-data> \
        datapath=<path-to-dataframe> \
        method=<method_name>

Our code supports 8 method_names:

1. AdaTriplet-AM : AdaTriplet loss with AutoMargin selection
2. AdaTriplet
3. Triplet-AM: Triplet loss with AutoMargin selection
4. Triplet
5. SCT
6. WAT
7. ArcFace
8. SoftTriplet

###Test

Run the script as follows:

  python test.py data_type=<OAI|CXR> 
  query_time=<time-in-months|all>
  pretrained_matching_model_folder_path=<path-to-folder-of-trained-models>

Results

Forensic Matching Results