TAPe+ML v3 Achieves High Performance with Under 100K Parameters
September 14, 2026
TAPe+ML v3 is a compact computer vision system using a structured Theory of Active Perception representation. It achieves 84.7 mAP50 on COCO object detection using fewer than 100,000 parameters by processing relations between perceptual elements instead of raw pixels.
HOW THIS AFFECTS YOU
●
builderYou can deploy highly capable vision models on edge devices with extremely limited parameter budgets.
●
researcherThis demonstrates an alternative to pixel-tensor-heavy architectures through structured perceptual encoding.