Automatic detection of Pneumonia from Chest X-rays
In this homework we will solve a real-life problem using deep learning. The objective is to build a model that can automatically detect pneumonia from chest X-rays.
The dataset is available at Kaggle.
The dataset contains 5,863 X-ray images (JPEG) and 2 categories (Pneumonia/Normal). Because of the large dataset I encourage you to test your code on a small subset of the data and then run it on HiperGator for the full dataset.
Specific objectives for this homework are:
- Practice creating a PyTorch dataset
- Practice creating a PyTorch data loader
- Apply what we have learned about CNNs to solve a real-life problem
- Practice using TensorBoard to visualize the training process
Submission
You need to create a folder inside the blue shared folder of your team. There you will upload your data
and code.
Download and visualize the data (10 pts)
Download the dataset from Kaggle and unzip it. The dataset is available at Kaggle.
Remove any additional files and folders that you do not need. The dataset should contain only the following folders:
train- training data with two subfoldersNORMALandPNEUMONIAtest- test data with two subfoldersNORMALandPNEUMONIAval- validation data with two subfoldersNORMALandPNEUMONIA
Make a notebook with a widget that allows the user to select a folder (train, validation or test)
and will visualize 4 or more random images from each of the two classes (NORMAL and PNEUMONIA).
Dataset (10 pts)
Create a PyTorch dataset MyGenerator.py that loads the data.
Inside your constructor you should receive at least the input_folder (train, validation or test)
and the transform to apply to each of the samples.
Tip: Remember that this dataset needs to return examples for both classes (NORMAL and PNEUMONIA). You can use a threshold on the received idx from the data loader to decide which class to return.
Tip 2: The images are very different in size, intensity and contrast. You can use transformations like: equalize, autocontrast, and resize to make the images more similar.
Model design (10 pts)
Please describe with words your proposed model architecture. Include information like which type of layers you will use, which activation functions to use for the intermediate and output layers, and the loss function to use.
Feel free to include an image of the proposed architecture.
One site for creating NN visualizations is NN-SVG.
Model training and tensorboard (10 pts)
Following our example from class, create a train.py script that will train your model. Class Example
In you training script you should:
- Iterate by a number of epochs and batches, and evaluate the model on the validation set at the end of each epoch.
- Create a tensorboard writer and save the following information for each epoch:
- Training loss
- Validation loss
- A sample of the input images, the ground truth labels, and the predicted labels
- Additionally, save the model graph for the first epoch
- Keep track of the model with the lowest validation loss and save it to disk.
Model evaluation (5 pts)
Create a main.py file or notebook that will have the general structure of a PyTorch project. Here you will call your dataset, data loader, model, and training script.
Describe results (10 pts)
Describe the results of your model. Include the following information:
- What was the best validation loss you got?
- What was the best accuracy you got?
- What was the best precision you got?
- What was the best recall you got?
- Show some of your results in tensorboard. Include a screenshot of the graph and the images. And discuss your results.