I’ve spent a good bit of time over the past few months learning a lot about Machine Learning and I thought I’d share some of that knowledge here. Let’s see, I started my Machine Learning journey sometime in March 2021 I believe and it’s currently the first of September of that same year, so let’s say it’s been a solid 6 months since I’d been tasked with creating an ML model that can recognize unique images on the sides of objects. Why I am being so vague is another story for another time. The problem was clear though, and I sorely underestimated the amount of research it would take to get my bearings on a solution. The solution I ended up landing on was Object Detection.
When first wading into the field of Object Detection, I quickly started to realize that everywhere I went to read about setting up an Object Detection model seems to assume you are a Data Scientist that already knows the nitpicky details. So I would have to jump from one article to another piecing together relevant bits of information until I finally had a grasp on what was going on. The first thing I remember was that Apple had a Machine Learning framework called CreateML that made it super easy to create models and that’s how I got my foot in the door.
So you went all in with Apple then right?
No, not hardly, I was able to train about 4 different classes of objects with about 100 images each which was really cool. However, this is when I realized that while training the models can be CPU / GPU time intensive, labeling the number of images that you need to create a useful robust model can be even more time intensive than that.
More about that tomorrow…
See ya later Buddies!
