YOLO26-L-obb
Ultralytics YOLO26 is the latest evolution in the YOLO series, engineered from the ground up for edge and low-power devices. This is the Oriented Bounding Box (OBB) variant optimized for detecting rotated objects in aerial and satellite imagery.
Model Specifications
| Property | Value |
|---|
| Input Size | 1024 pixels |
| mAP Test (50-95, e2e) | 56.2 |
| mAP Test (50, e2e) | 81.6 |
| CPU Speed (ONNX) | 735.6 ms |
| T4 TensorRT10 Speed | 13.0 ms |
| Parameters | 25.6M |
| FLOPs | 230.0B |
Key Features
The architecture of YOLO26 is guided by three core principles:
Simplicity: YOLO26 is a native end-to-end model, producing predictions directly without the need for non-maximum suppression (NMS). By eliminating this post-processing step, inference becomes faster, lighter, and easier to deploy in real-world systems.
Deployment Efficiency: The end-to-end design cuts out an entire stage of the pipeline, dramatically simplifying integration, reducing latency, and making deployment more robust across diverse environments.
Training Innovation: YOLO26 introduces the MuSGD optimizer, a hybrid of SGD and Muon — inspired by Moonshot AI's Kimi K2 breakthroughs in LLM training. This optimizer brings enhanced stability and faster convergence.
Refined OBB Features
- Specialized Angle Loss: Improved detection accuracy for square-shaped objects
- Optimized OBB Decoding: Resolves boundary discontinuity issues
- DFL Removal: Simplified inference and broader hardware compatibility
- Up to 43% Faster CPU Inference: Optimized for edge computing
Usage
Install ultralytics with pip install ultralytics.
Download the model.
1from huggingface_hub import hf_hub_download
2
3model_path = hf_hub_download(repo_id="openvision/yolo26-l-obb", filename="model.pt")
Infer.
1from ultralytics import YOLO
2from PIL import Image
3import requests
4
5model = YOLO(model_path)
6
7url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
8image = Image.open(requests.get(url, stream=True).raw)
9
10# Run inference with the YOLO26l-obb model on the image
11results = model.predict(image)
Documentation
For more information, see the
official YOLO26 documentation.
