Open-source shopper attention heatmap: YOLO11-pose + Gaze-LLE + Depth Anything V2

claudioe · reddit · 2026-10-16

Inspired by CactusXR (Auki)'s retail attention heatmaps, a developer built a similar system from a public convenience-store CCTV video using only open-source models: YOLO11-pose for person/keypoint detection, Gaze-LLE for gaze direction estimation, and Depth Anything V2 for depth and 3D scene reconstruction. The pipeline accumulates gaze frame-by-frame into shelf attention heatmaps and ranks store zones. Caveats: not real-time, no ground-truth validation. Code and demo video are public.

Original post →

More from Apps

Apps channel →