search

THU-MIG / yolov10

YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]
https://arxiv.org/abs/2405.14458
GNU Affero General Public License v3.0
9.9k stars 978 forks source link

YOLOv10 vs YOLOv5 Inference Performance Test on Horizon J5 Chip (640x640 Input) #127

Open 315386775 opened 5 months ago

315386775 commented 5 months ago

Performance Metrics

Model Test Size #Params FLOPs APval HorizonJ5 Latency HorizonJ5 Latency
YOLOv5-S 640 7.2M 16.5G 37.4% 3.18ms 314
YOLOv10-S 640 7.2M 21.6G 46.3% 2.05ms 488

Test Environment

Chip: Horizon J5.

Details and Reproduction Steps:

  1. Convert onnx model

yolo export model=yolov5s.pt format=onnx opset=11 simplify yolo export model=yolov10s.pt format=onnx opset=11 simplify

  1. Check model and HorizonJ5 Test

hb_mapper makertbin --config yolov10/5.yaml

Yolov10 is a great work, those who have efficiency requirements can start using it now, thks!

benx13 commented 5 months ago

can you benchmark yolov8n and yolov8s as well?