The repository benchmarks various pre-trained models to reveal the best performing model for the task of lung disease classification. The used dataset: ChestX-ray14, which consists of 112k 1024x1024 pixel x-ray images of 14 distinct lung diseases as well as images of healthy individuals.
What the preprocessed images look like:
For using the models without re-training: 1 GB CUDA and 1 minute of execution / no CUDA and 5 minutes of execution.
For using the models with re-training: 10 GB CUDA