What is Deep Runner?


01. What is Deep Runner?


Deep Runner is a general purpose visual sensor device that recognizes objects in an image using deep learning technology.

02. Deep Runner Products



03. Supporting Deep Learning Algorithm and Recognition Rate


  • The Deep Runner recognizes and executes correctly with the recognition rate of the world-known deep-running algorithm (SSD300, Inception v1, etc.). [Human brain = deep learning algorithm]
  • Due to the nature of the deep learning technology, the correct recognition rate changes according to the user's learning of an object with an image, no matter how good the algorithm is. [What the user should do = Collect various image data of the object to be recognized]
  • Provide customers with firmware upgrade files. Supported Deep Learning algorithms will be added at all times.



04. Function


Deep runner supports Classification algorithm and Detection / Location algorithm. Select the algorithm according to the nature of the application you want.


  • Classification : Tells what the image is about.
  • Detection/Location : Tells where objects are.
  • Multiview : Up to 16 channels per Deep Runner can be recognized simultaneously.



05. What objects can be recognized?

  • It can recognize various categories such as gender, age, emotion, clothing as well as all objects with characteristic. (* Individual facial recognition is not supported at present.)
  • Deep Runner can also be recognized if they are visible and distant from the screen. However, it is possible to distinguish between size and length when camera zoom function input to Deep Runner and object position are fixed.



06. Advantages of Deep Runner compared to existing GPU based Deep Learning System

  • The existing GPU-based deep learning system requires a large volume, high cost and power for the GPU itself in the computer, and requires an additional deep learning development environment so that general developers, not deep learning experts, There is difficulty.
  • However, Deep Runner is a small, low-power module implemented in FPGA rather than GPU. It is designed to be able to calculate deep learning algorithm in module itself and low cost compared to GPU, so users do not need further development in deep learning recognition.
  • It is universally applicable as a part in the part where deep running image recognition is required in any large system.



For more technical information, see the Deep Runner User Manual.