/da-encoding

Primary LanguagePythonMIT LicenseMIT

da-encoding

Repo with code for upcoming paper proposal.

Installation

Simply run

pip -r requirements.txt

Usage

First, run train_resnet18_with_aug.py with desired augmentations (e.g. uncomment the ColorJitter augmentation). The checkpoint will be generated in a new folder called lightning_logs. Copy the checkpoint path to load_backbone in ranking_and_backbone.py, then choose the desired augmentation (e.g. generate_brighted_imgs) and run that file. Checkpoint will again be in lightning_logs. Copy it to layer_relevance.py and it will print the results.

A more automatized workflow might be avaiable once the paper is finished.