Some test code about using CNTK framework's evaluate performance, compare with OpenCV DNN module using caffe model.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
CntkEvaluatePerformance
PythonCntk
PythonOpencvDnn
packages/CNTK.GPU.2.2.0
CntkEvaluatePerformance.sln
README.md

README.md

CNTK Evaluate Performance Test

This code is some test about using CNTK framework's evaluate performance, compare with OpenCV DNN module using caffe model.

Test environment

Framework

System

  • Windows 10 Pro 64bit

Hardware

  • Intel Core i7-7820HQ @ 2.90GHz
  • nVidia Quadro M1200

Test Model

Deep Residual Networks a.k.a ResNet, this test choose ResNet50_ImageNet pre-trained model to test.

Test Image

Space Shuttle
Space Shuttle

Project Introduction

In this solution, we have 3 projects for 3 diffetent test.

  • CntkEvaluatePerformance : Evaluate with CNTK in C#.
  • PythonCntk : Evaluate with CNTK in python.
  • PythonOpencvDnn : Evaluate with OpenCv DNN in python.

Most code are almost follow example to keep it simple.

Test Result

  • CNTK in C# : 1117ms CNTK C# Result
  • CNTK in Python : 1114ms CNTK C# Python
  • OpenCv DNN Caffe model in Python : 151ms OpenCv DNN Result

Evaluate one image in CNTK does 7 times slower than evaluate in OpenCv DNN, Why ?