// the find
waldo-vision/optical.flow.demo
A project that uses optical flow and machine learning to detect aimhacking in video clips.
An abandoned attempt to detect 'closet hacking' (humanized aim-assist) in FPS gameplay clips by analyzing mouse movement patterns via optical flow and deep learning, rather than scanning the client. Interesting problem framing — it's for anti-cheat researchers or ML hobbyists curious about motion-based detection, not for anyone wanting a working tool.
The core idea is sound: aim-assist produces statistically different acceleration curves than human input, and optical flow is a reasonable way to extract that signal from video alone, no client access needed. The repo ships both Farneback (dense) and Lucas-Kanade (sparse) optical flow implementations, so you get two different motion-estimation approaches to compare against each other.
Last commit is from November 2021 — this is a dead project, not actively maintained. There's no actual classifier: no model code, no training data, no accuracy numbers, just the two optical flow scripts with nothing downstream turning motion vectors into a cheat/no-cheat decision. The README is a pitch deck (vision, skills needed, a recruiting joke) rather than documentation — no setup instructions, no example input/output, no results from the 'phase 1' work it claims is in progress.