STREAM: Stochastic Riemannian Flow Matching with Anisotropic Decoder for Digital Histopathology Image Generation
STREAM generates synthetic histopathology images using a Riemannian flow matching approach on Vision Foundation Model features. It overcomes common model failures by using a stochastic bridge to ensure smooth transitions and an anisotropic decoder informed by Jacobian analysis, leading to state-of-the-art reconstruction and generation performance on cancer datasets.