Project 02 — Research

DPC-YOLO26

NMS-free adversarial-patch defense. Turns diffusion residuals into a suspicion field that suppresses patch detections while retaining real objects.

Python PyTorch Diffusion Capstone

Overview

DPC-YOLO26 is a defense against adversarial-patch attacks on NMS-free object detectors. It turns multi-timestep diffusion residuals into a suspicion field that suppresses patch detections while retaining real objects.

It's independent research, and it was my cybersecurity capstone. The full write-up is on the Research page, and the code is on GitHub.