Skip to content

Latest commit

 

History

History
33 lines (25 loc) · 1.52 KB

README.md

File metadata and controls

33 lines (25 loc) · 1.52 KB
Raw

MobileNetSSD Caffe-OpenVino-comparison

This project indicates performance differences of using MobileNetSSD with Caffe and OpenVino

- it does not contain the used video for legal reasons as it is property of Udacity

Caffe vs. OpenVino

(size, execution-time, CPU-usage)

Video replay without model

  • size: 0 KB (no model used)
  • execution-time: 2 minutes 24 seconds
  • CPU-usage (avg of 3 executions): 3.70%

Caffe-Model

  • size: MobileNetSSD_deploy.caffemodel: 23.1 MB, MobileNetSSD_deploy.prototxt: 29 KB
  • execution-time: 3 minute 9 seconds 2:48
  • CPU-usage (avg of 10 executions): 9.97%

(Note: Original-Code was retrieved from: pyimagesearch)

OpenVino

  • size: MobileNetSSD_deploy.bin: 23.1 MB, MobileNetSSD_deploy.xml: 175 KB
  • execution-time (speed): 4 minutes 22 seconds
  • CPU overhead (avg of 10 executions): 8.14% (-1.83% vs. default Caffe-Model)

Comparison-Conclusion

Only slight improvement was noticed when using the model converted to IR with OpenVino, when it comes to CPU-usage, which decreased by 1.83%. Those results were expected as MobileNet models are already well optimised for speed and size.

Running instructions

  • Video replay without model: Use the running instructions mentioned at the end of simple_video.py
  • Caffe-Model: Use the running instructions mentioned at the end of detection_caffe.py