motivation
directory contains code used to create the example in Section 2: Motivation.
scripts/lightning_trainer.py
trains models on CheXpert dataset.scripts/radimagenet_pretraining.py
trains models on RadImageNet dataset.scripts/finetuning_with_masks.py
fine-tuning of models on CheXlocalize dataset with masks.scripts/create_train_val_test_split.py
creates train/val/test splits for CheXpert dataset.scripts/chexlocalize_finetuned_heatmaps.py
generates explanation heatmaps for (mis)aligned models.scripts/chexlocalize_heatmaps.py
generates explanation heatmaps for trained models.
notebooks/linear_regression_localization_accuracy_analysis.ipynb
notebook in which analysis of effects was performed.notebooks/creating_dataset_for_effect_analysis.ipynb
notebook in which dataset for analysis was created.notebooks/generate_explanations_plots.ipynb
notebook used for creation of explanation plots and example images with masks on CheXlocalize.
results
directory contains metadata and code used to create figures and tables in Section 4: Experiments.