BigCell utilizes a flow-matching transformer trained at 2048x2048 pixels on 120,000 WSIs to generate coherent, gigapixel-scale histopathology images. It conditions generation on embeddings from pathology foundation models to bridge the gap between patch-scale models and whole-slide diagnostic labels.
HOW THIS AFFECTS YOU
●
researcherYou can generate high-resolution, slide-scale data for training pathology models.
●
healthYou can leverage more realistic synthetic tissue scans for diagnostic research and training.