Evaluation of explainability methods and robustness in image classification

• fine_tune.py for fine tuning selected model to COCO dataset via multilabel classification. Currently supports ViT, VGG and ResNet models from torchvision pretrained models.
• shape_robustness.py contains script to evaluate shape robustness of fine tuned model.
• visualize_examples.py contains script to visualize random examples of images as how they would show in shape robustness evaluation.
• get_image_to_annotation.py has script that generates json in form {image_id: annotation id}, which contains all valid images and their chosen annotations. Selection criteria takes largest segmentation mask that is at least 75% within the center square of the image.
• utils.py contains Pytorch dataset implementations for training and shape robustness evaluation.