Skip to content

Latest commit

 

History

History
64 lines (56 loc) · 2.18 KB

covidnet_pneumonia.md

File metadata and controls

64 lines (56 loc) · 2.18 KB
Raw

COVIDNet-P: COVID-Net for Pneumonia Detection

This section describes how we leveraged COVID-Net to create a COVIDNet-P model which distinguishes between cases where pneumonia is present and where it is not. Using the inference_pneumonia.py file provided, COVIDNet-P takes as input an image of shape (N, 480, 480, 3) and outputs the likelihood of whether pneumonia is present or not in the given image.

Steps for Inference

DISCLAIMER: Do not use this prediction for self-diagnosis. You should check with your local authorities for the latest advice on seeking medical assistance.

  1. Download a model from the pretrained models section
  2. Locate models and xray image to be inferenced
  3. To inference,
python inference_pneumonia.py \
    --weightspath models/COVIDNet-CXR4-A \
    --metaname model.meta \
    --ckptname model-18540 \
    --imagepath assets/ex-covid.jpeg
  1. For more options and information, python inference_pneumonia.py --help

Steps for Evaluation

  1. Download a model from the pretrained models section
  2. Download a test dataset from the main repo
  3. To evaluate,
python eval_pneumonia.py \
    --weightspath models/COVIDNet-CXR4-A \
    --metaname model.meta \
    --ckptname model-18540 \
    --testfile test_COVIDx5.txt
    --testfolder data/test
  1. For more options and information, python eval_pneumonia.py --help

Results

These are the results generated using the eval_pneumonia.py script with the COVIDNet-CXR4-A model acting as the backbone of COVIDNet-P and test_COVIDx5.txt dataset

Sensitivity (%)
Normal Pneumonia
94.0 95.0
Positive Predictive Value (%)
Normal Pneumonia
90.4 96.9