/StrepDetection

An interpretable deep learning approach to detect strep throat directly from cell phone videos.

Primary LanguageJupyter Notebook

StrepDetection

Detection of strep throat directly from cell phone videos.
Employing intermediate symptom classification combined with rule-based decisions for interpretable results.
Implementing strategies (hard-negative mining, contrastive learning) to combat limited and imbalanced data.

Parsing data from CVAT:

  1. Download data from CVAT
    • Actions > Export Dataset > Export Format: CVAT for video 1.1.
    • This will download a folder containing an xml file with the dataset annotations.
  2. Parse annotations via parse_xml.py
    • Set the xml file path and run parse_xml.py.
    • This will produce a .csv file with the video, frame, and relevant labels.
  3. Merge CVAT data with .xlsx data
    • Follow the steps in data_process.ipynb.
    • This will merge the annotations from the .xlsx training review with the CVAT labels, checking for any overlap.

Model Checkpoints:

OneDrive folder containing model checkpoints.

Authored by Rishi Chandra, rchand18@jhu.edu, as part of the ARCADE Lab at Johns Hopkins University.