besit/yolo-onnx

YOLO26 ONNX — Detection Models

This repository provides YOLO26 detection models exported to ONNX format, for use with ONNX Runtime or any ONNX-compatible inference engine.

These models are not my work. YOLO26 was developed by Ultralytics and is distributed under the AGPL-3.0 license. I am sharing these pre-exported ONNX versions as a convenience for those who prefer not to handle the export step themselves.

---

Models

Variant	File
Nano	yolo26n.onnx
Small	yolo26s.onnx
Medium	yolo26m.onnx
Large	yolo26l.onnx
XLarge	yolo26x.onnx

All models were exported from the official Ultralytics .pt checkpoints using:

yolo export model=yolo26n.pt format=onnx opset=18

Input shape: [1, 3, 640, 640] — normalized RGB, values in [0, 1].

Output shape: [1, 300, 6] — 300 detections × (x1, y1, x2, y2, confidence, class_id), in xyxy format, NMS-free.

---

Attribution

• Model: YOLO26 by Ultralytics
• License: AGPL-3.0
• Original weights: ultralytics/assets
• Export: ONNX opset 18, no modifications to weights or architecture

If you intend to use these models in a commercial product, you will need a commercial license from Ultralytics. See ultralytics.com/license.

---

Why this repo exists

Exporting YOLO26 to ONNX requires a working Python environment with Ultralytics installed. This repo saves that step. If you prefer to export yourself, follow the official Ultralytics export guide.

Converted using Ultralytics tools.