EDGE//YOLO26
Portfolio
RUNTIME ·
onnxruntime-web
Real-time object detection · runs entirely in your browser

Detection at the edge

This YOLO26 detector is NMS-free and trained on dense retail shelves. It counts products without sending a frame off this device. There is no upload, server, or API. The page runs the same ONNX graph locally with WebGPU.

0.572mAP@50-95
640²input
NMS-FREEend-to-end head
SKU-110K~150 objects / image
Initialising detector
connecting…
PICK A SHELF BELOW · OR DROP AN IMAGE
DETECTION RUNS ON-DEVICE
Inference latencyidle
ms
Throughput
fps
Objects
Model card
architectureYOLO26-s
params~9.6 M
datasetSKU-110K
precisionFP32
output[1 · 600 · 6]
The image never leaves your machine. Preprocessing, inference, and box decoding all happen in this browser tab through WebGPU.