Synthetic Data Augmentation

This repo accompanies the FF25 research cycle focused on exploring the impact of synthetic data augmentation on tasks associated with the Severstal Steel Defect Detection dataset hosted on Kaggle.

Downloading the data

Register an account on Kaggle, and download and unzip the data into the data subdirectory of this repo. For most of the notebooks and scripts in this repo to work, the resulting directory structure should be:

data
├── sample_submission.csv
├── severstal-steel-defect-detection.zip
├── test_images
├── train.csv
└── train_images