PIGMENT

PIGMENT: A Deep Learning Framework for Porcine Immunohistochemistry Segmentation Undergraduate Researcher, Advisor: Prof. Ragini Verma - Built PIGMENT, a SegFormer-B0 pipeline with APP-specific augmentation (copy-paste, GMM label promotion, morphological bridging) on 525 expert-annotated tiles from 3 pigs, achieving 0.85 detection rate and 0.74 Dice score. - Outperformed UniverSeg and MicroSAM baselines by 67% and >95% on detection rate and 5.7x on Dice; boosted detection rate 5% using 31% fewer tiles via cross-animal training composition.