Why you should monitor your pictures’ sharpness when deploying Computer Vision models

By Medium - 2021-03-22

Description

But if you place the frog into cold water and gradually increase the heat, the frog will enjoy this hot bath for a while and, before long, it will unresistingly allow itself to be boiled to death. We…

Summary

  • Permanently assessing the quality of the pictures you are getting from a device might help you keep a robust model instead of a blind one!
  • I wish I had been able to share the whole picture but I cannot for obvious confidentiality reasons) As this section is used by the model to infer some characteristics of a manufactured product, needless to say than the performance of the Deep Learning algorithm starts to decrease slowly as the picture gets blurry, which might be harder to detect than going from a sharp picture to a blank one (Remember the frog story?).
  • So, in this particular case, it might be interesting to warn shopfloor people when the picture provided by the camera starts to go beyond a pre-defined threshold of sharpness and requires cleaning.
  • Gradient calculation of the [1, 3, 0] vector — Image by Author From 1 to 3, the function is “y = 2x”, its derivative being “2”.

 

Topics

  1. Machine_Learning (0.4)
  2. Backend (0.21)
  3. NLP (0.17)

Similar Articles

Making concurrent API calls in Node

By DEV Community - 2020-12-27

The problem When building up a backend API, it is common that we need to fetch data from a... Tagged with node, javascript.

Ensemble Learning: Stacking, Blending & Voting

By Medium - 2020-12-13

We have heard the phrase “unity is strength”, whose meaning can be transferred to different areas of life. Sometimes correct answers to a specific problem are supported by several sources and not…