YOLOv12 is an open source real-time target detection tool led by GitHub user sunsmarterjie, belonging to the latest evolution of the YOLO (You Only Look Once) series. The project is jointly participated by researchers from the University of Buffalo and the Chinese Academy of Sciences, focusing on high-precision, low-latency target detection through deep learning techniques.
Core improvements include:
- Attention mechanism optimizationIntroducing Area Attention and R-ELAN module to automatically optimize the feature extraction process and significantly improve the detection capability of small targets.
- multimodel architectureProvides five types of pre-trained models, Nano/Small/Medium/Large/Extra-Large, adapting to different hardware environments from edge devices to server clusters.
- Accuracy-Speed Balance: On T4 GPUs, the base model (YOLOv12-N) reaches 40.61 TP3T mAP while maintaining ultra-low latency of 1.64ms
- Deployment Friendliness: Supports export to ONNX and TensorRT formats for easy deployment in industrial-grade applications
The project adopts the GNU AGPL-3.0 open source protocol, which completely retains the advantages of the YOLO series single-stage detection architecture, while reducing the complexity of the algorithm through modular design.
This answer comes from the articleYOLOv12: Open source tool for real-time image and video target detectionThe
May not be reproduced without permission:AI productivity tools " What is the YOLOv12 and what are its core improvements over the traditional YOLO series?
