Crunch is a python package that generates data visualizations on large image datasets to enhance the pre-training process. The objective of Crunch is to merge explainable AI with beautiful data visualization into a painless python library.
Crunch works by extracting embeddings (feature extraction is customizable), reducing the dimensionality of those embeddings, and applying classic clustering algorithms (e.g. k-means clustering, HDBSCAN, etc.) to create interpretable graph visualizations. Here is a high-level diagram detailing the pipeline and underlying infrastructure behind Crunch:
