Shrey Viradiya

Pneumonia Detection using NVIDIA DALI

May 24, 2024

This repository contains code for pneumonia detection using X-ray images of the lungs. Test accuracy achieved (highest) : 96.5 %

Github Repo Link

Link for the dataset

Link to the trained models

Pneumonia Detection with NVIDIA DALI

 

Instructions

  • Install required packages from requirements.txt

 

  • You can download, extract and move the data according to the label using data.py
    • pip install kaggle
    • Download Kaggle API key from your Kaggle Account. Go to www.kaggle.com -> My Account -> Create New API token
    • Place the file in on your home directory $(HOME)/.kaggle/kaggle.json.
    • Kaggle module will look for this token at ~/.kaggle/kaggle.json on Linux, OSX, and other UNIX-based operating systems, and at C:\Users\<windows-username>\.kaggle\kaggle.json on Windows.
    • execute python data.py

 

  • Run main.py to train from the dataset. For example, python main.py --base_model {base_model} --optimizer {optimizer} --learning_rate {learning_rate} --batch_size {batch_size} --epoch {epoch} --nvidiadali --colab
    • To change pretrained base model, give input while initializing the model object. Use values from
      • ResNet18
      • ResNet34
      • ResNet50
      • ResNet101
      • ResNet152
      • Alexnet
      • VGG11
      • VGG13
      • VGG16
      • VGG19
      • GoogleNet
      • Inception
    • If using colab for training, mount the drive and use —colab to save the files in the drive
    • For faster training and Image augmentation, if on Linux, NVIDIA DALI can be used
      • Install NVIDIA DALI. Please see the installation page for details.
      • Use —nvidiadali flag to use it
      • Note that NVIDIA DALI is only available for Linux
    • To change optimizers, use one of the following
      • Adam
      • SGD
      • RMSprop
      • Adagrad
      • Adadelta

 

  • To test new dataset, run test.py with directory path and base model with option given above. Make sure that model is trained on those models first.
    • Example, python3 test.py "./data/Corona_Classification_data/test/" Inception True True

 

  • To generate a Class Activation Map from a trained model, after training it, use CAM.py like python3 CAM.py {Model_Name} {Path_to_Image}
    • Example, python3 CAM.py Inception "./data/abc.jpg"
    • This will save the output as CAM_{Model_Name}.jpg in the current directory