AI BriefingDudaji Utilizes Warboy for Workplace Safety Monitoring
Key point
AI service company Dudaji built a real-time workplace safety monitoring system using FuriosaAI's Gen 1 Vision NPU.
Details
AI service company Dudaji built a real-time workplace safety monitoring system using FuriosaAI's Gen 1 Vision NPU (Warboy). This system is used to rapidly detect complex safety issues such as whether workers are wearing protective equipment, access to hazardous zones, and fire detection.
Dudaji ported a YOLO-family object detection model to the NPU, then exported the model via ONNX and optimized it through Calibration and Quantization processes. The key performance metrics are as follows.
- Based on the YOLOv5s model, achieved 40 FPS when running 8 streams simultaneously
- Recorded a peak performance of up to 330 FPS on a single NPU card
The Furiosa SDK is compatible with major AI frameworks such as PyTorch and TensorFlow Lite, and allows models to be converted to ONNX format for quantization using ImageNet samples. In addition, the NPU is provided through Kakao Cloud, offering an environment where users can remotely access instances for testing.
During the optimization process, Dudaji tested various quantization algorithms and confirmed that the MinMax calibration method was the most effective for the dataset.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Rebellions Accelerates Development Tasks Including Physical Design by Building AWS-Based EDA Environment