The Problem

You’ll notice all of my project write-ups start with identifying the problem my solution solves. This write-up summarizes each of the interactive articles and tutorials available in the articles section of this website, and so the problem being solved is a bit more intanbigle/conceptual.

By making content interactive, it encourages a higher level understanding from the reader, and requires a higher level communication ability from the creator. Therefore, I consider the problem that interactive content solves to be communication shortfalls that are common in technical content. Whether that be from a lack of understanding/focus from the consumer, or a lack of ability to communicate from the creator. Communication is everything.

A secondary objective of this content is to fill-in any gaps I see in other publicly available content. Companies such as roboflow have enough content such that any diligent person with a basic programming ability could become a half-decent computer vision engineer. My goal is to provide original content focused on real-world use cases.

 

Quantifying Focus in Computer Vision Use Cases

Audience: Computer Vision Engineers, especially those interacting with dataOps, developing sensors, or any data-centric developer.

Topic: Image clarity, or lack thereof, is critical for many reasons in any computer vision project. It’s possible to quantify clearity.

Interactive Content:: The Colab Notebook downloads a dataset from roboflow, implements brenner’s focus measure and 4 other variations, evaluates each variations effectiviness on the given dataset, and then simulates a threshold based filter (dropping blurry images). Success is measured by the number of blurry images the filtering logic can successfully identify and drop.

Conclusion: Relative performance of a given focus measure is dependent on the dataset and use case. However, implementing focus measure operators is not a novel task and so finding an effective focus measure is closer to a routine experiment than novel task.