Instruction-Conditioned Pathology Segmentation
CIPS-Net is my histopathology segmentation work focused on instruction-driven, compositional reasoning for overlapping pathology classes. The goal is to move beyond fixed-label segmentation and make the model respond to human-readable pathology instructions.
The project combines vision-language guidance with graph reasoning to improve dense prediction on pathology data. It connects directly to my broader research on medical imaging, multimodal learning, and clinically useful AI systems.