Release Notification of Deep Runner

30 Nov 2018
Views 2078

Dear customers,

Thank you for purchasing and using the Deep Runner, AI visual sensor device

We are pleased to upgrade our Deep Runner and Deep Trainer to new versions.

Upgrade includes:

  • Deep Runner firmware: 20181116_SSD version
  • Deep Trainer: version 1.4

The updated file can be downloaded from the link below.

To check for new features, please upgrade your Deep Runner and Deep Trainer and follow Sections 2.30, 2.31, and 2.32 of Deep Runner Tutorial document.

Features of the new upgrade

  • Support for MobilnetV2_1.4 classification algorithm (For quality inspection applications)
  • Includes easy-to-use TF package for training SSD object recognition algorithm using Tensorflow
  • Support for full HD (1080P 60Hz) video input (Applied only for Deep Runner sold after September 2018)
  • Other improvements

New Mobilnet V2_1.4 Deep Learning Classification Algorithm

  • The MobileNet V2 1.4 is an algorithm released by Google in April 2018 that has 1.4 times more parameters than the MobilnetsV2 1.0 and has better recognition capability than GoogleNet, VGG16, and NasNet (Mobile). The Deep Runner runs the algorithm at a high speed of 60 frames per second (compared to 20-27 frames per second for GoogleNet).
  • The existing classification algorithm (GoogleNet) and the object recognition algorithm (SSD) are supported by different firmware, and the firmware has to be changed to use alternately. However, it is not necessary for the MobilenetV2 and SSD because they are supported by the same firmware.
  • The training of new MobinetV2 algorithms is supported by Deep Trainer software, and the Deep Trainer’s training results are accurately reflected in Deep Runner devices, in contrast to existing GoogleNet algorithm where the results of training and the results of the Deep Runner did not exactly match. Therefore, optimizing the Top1 training results in the Deep Trainer will ensure the best recognition rate in the Deep Runner.

Easy-to-use Tensorflow support for training SSD object recognition algorithm

  • SSD can be trained by Google's open source software Tensorflow, but it is very complicated and inconvenient to use. This Deep Trainer contains TF package with which you can train SSD and MobilenetV2 algorithms with a simple procedure on your Windows PC.
  • The SSD algorithm we tested using the Tensorflow can be used effectively in factory quality inspections as well as security systems, smart car applications.
  • For more information, please refer to Section 2.31 of the Deep Runner Tutorial v1.4 document.

How to Upgrade

  • For instructions on upgrading the Deep Runner, please refer to sections 2.25 and 2.26 in the Deep Runner Tutorial document.
  • To upgrade the Deep Trainer, first uninstall the existing Deep Trainer on your PC (if any), and refer to Section 2.25 and 2.26 of the Deep Runner for installation instructions.


Neurocoms Inc.