NexaAI/yolov12-npu

YOLOv12-N

Model Description

YOLOv12-N is a variant of the 12th-generation YOLO (You Only Look Once) real-time object detector.
It builds on prior YOLO models with improved backbone/neck architectures, updated training strategies, and optimizations for both high-performance GPUs and edge devices.

Features

• Real-time object detection optimized for low-latency inference.
• High accuracy across diverse categories and challenging environments.
• Lightweight variants suitable for mobile and embedded deployment.
• Scalable: runs from smartphones to multi-GPU servers.
• Extensible: fine-tuning supported for domain-specific datasets.

Use Cases

• Autonomous driving and ADAS (Advanced Driver Assistance Systems)
• Surveillance and security monitoring
• Industrial automation and defect detection
• Retail analytics and inventory monitoring
• Sports analytics and event detection

Inputs and Outputs

Input:

• RGB images or video frames (any resolution; auto-resized during preprocessing).

Output:

• Bounding boxes (x, y, w, h)
• Class labels
• Confidence scores

---

How to use

⚠️ Hardware requirement: the model currently runs only on Qualcomm NPUs (e.g., Snapdragon-powered AIPC).
Apple NPU support is planned next.

1) Install Nexa-SDK

• Download and follow the steps under "Deploy Section" Nexa's model page: Download Windows arm64 SDK
• (Other platforms coming soon)

2) Get an access token

Create a token in the Model Hub, then log in:

nexa config set license '<access_token>'

3) Run the model

Running:

nexa infer NexaAI/yolov12-npu

---

License

• Licensed under AGPL-3.0 (same as Ultralytics YOLO).

References

• Original repo: https://github.com/sunsmarterjie/yolov12
• Ultralytics YOLO family: https://github.com/ultralytics/ultralytics