But wisdom is justified of all her children.
— Luke 7:35
Anything but another demon pic, every one of my Facebook, Twitter, Google+, Instagram, Path, LinkedIn and Swarm followers have thought at one point or another–if not outright said. Are you obsessed with demons or something? all have thought—if not outright said.
I've always ignored these questions in the past, as they are so stupid to ask, and too stupid to answer; but, if I had answered them, I would have said, I hear ya, and, No, respectively.
Ironically, insodoing, I would have begged what I consider to actually be a valid question: Then, what's the deal?
I wouldn't have answered that one, either, though, until I finished that which is presented in this post. The answer would have sounded crazier than taking pictures of violent demons sounds; but, now that I am finished—and alive, thank you very much—I will answer that question:
Introducing the world's first and only machine-learning image classifier model for automatic and instant recognition of demons in digital media:
 |
| Registered Apple Developer Connection members can download the first version of the world's first and only demon-recognition CoreML Model (17 Kb) to integrate into their iOS 11.4 (or higher) camera or video apps |
To make the model required lots of demon pics. Fortunately, lots of demons were available for recording on digital video. I posted hundreds of them over the past several months to my social media accounts as I acquired and processed them [see Social Media Ad Promos (2017-2018) on Behance.net]. As of yesterday, I fed about 100 of them into Apple's CreateML playground per the instructions in Creating an Image Classifier Model posted to the Apple Developer Connection site.
I got the idea to create the model after integrating Core Image's facial recognition plug-in into an experimental version of my upcoming app, Chroma. Although the framework supplied by Apple for integration by developers purports to recognize only human faces, it also captured some demon faces, too—albeit less than half. It stood to reason that a model trained with only demon faces would perform even better.
The process for creating and testing the model was incredibly simple, although preparing that many images for doing so was a bit tedious. Each image used for creating had to be cropped to just the demon's face only:
Once the images were cropped, they had to be categorized. To start, I separated the individual demons that sometimes group to create the facade of a much larger demon (they look like birds with outspread wings) from images showing actual larger demons (like the one shown in the video above). Then, using the Create ML developer tool in an Xcode playground, I trained and tested my image classifier model:
NOTE | Ever since Google shutdown my YouTube account (without warning and without reason—and without remedy), I've been uploading videos directly to the posts versus hosting them on YouTube; for whatever reason, some look okay (like the first video), while some don't (like the video above). You know, they sure do get a lot of investor money; what are they spending that on, exactly?
The following screen shot shows the statistical (and theoretical) end result of my efforts:
 |
| Version 0.1 of my image classified model identifies the smaller, bird-like demons that sometimes stack on top of each other to form a single (assembled) demon 98.91% of the time, whereas it recognizes an assembly of them 93.75% of the time; the model was trained with only 100 out of literally ten of thousands of images showing demons |
 |
| Stubborn stupidity is a cross no one should bear under the circumstances |
