Convolutional Neural Network (CNN) Image Classifier

Document created by edk Employee on Jul 19, 2017Last modified by edk Employee on Jul 19, 2017
Version 2Show Document
  • View in full screen mode

Demo

Neural network classification method based on SqueezeNet. Images are captured by the camera processed and classified by the S32V processor and then displayed on the TV monitor with a confidence percentage calculated for each object visualized.

 

  • Based on SqueezeNet, 501x fewer parameters than AlexNet
  • Low power consumption - Less than 10 watts total
  • Average top 1 accuracy of 58% and top 5 accuracies of 92%
  • CNN built with APEX-DNN library

 

Product

Attachments

    Outcomes