Swin Transformer with CutMix for Technosignature Detection in the SETI Kaggle Challenge on Green Bank Telescope Spectrograms

To the point

An ML competition to detect technosignature signals in Green Bank Telescope 2D spectrogram slices by training a Swin Transformer with transfer learning, data augmentation, and a weighted sampler to handle class imbalance, using mixed precision and test-time augmentation, run with HuggingFace Accelerate and Weights & Biases, and following a simple workflow to download data, install dependencies, train with main.py, and evaluate with evaluate.py to generate submissions (often seconds per run, with an example of five epochs).