Representation learning on PUNCH data

We use self supervised learning on PUNCH imagery that generates dense vector representations known as embeddings. We reuse those embeddings for multiple applications namely artifact correction, image search and CME detection.