hheiden/dinov2-mist-molecular-depiction-contrastive

A DINOv2 ViT-S/14-reg fine-tuned so that an image of a molecular structure diagram embeds where its molecule embeds in the frozen MIST-28M embedding space. Objective: smooth-L1 regression onto the frozen target, no negatives (the JEPA move).

# GigaHeart ## A Cardiac-specific CT Foundation Model for Heart Transplantation

Directed message passing neural networks for property prediction of molecules and reactions with uncertainty and interpretation.