Labels and images have a very important relationship in machine learning models, as it is of utmost importance to the functionality of the models. There have been many vocalized concerns of biases in the labelling of images in recent years of machine learning. Things like racially charged classifications, biased datasets and cultural differences in definitions can cause serious problems when allowed to exist in what is supposed to be standardized functions of massively popular products like Googles facial recognition or online chatbots. Matching labels to images can be surprisingly difficult, especially when the labels are man made definitions like “engineer” as opposed to something completely standardized like a “cocker-spaniel”.
I created a model that will tell if my phone is plugged in, and alert me if it is not. https://editor.p5js.org/liamtsang/sketches/L4BVwW53x
I found the creation of this to be a bit more difficult than I expected, the model seemed to get confused if the screen was on or off, and the camera angle had to be very exact or it wouldn’t work. I tried to create the same idea earlier in class, but my laptop camera had too wide of an angle to be consistent. This setup at home with an external webcam worked far better and more consistently. I came up with this idea because I had forgotten to charge my phone the day of class last week and thought this could be a fun solution. The second go around definitely was easier as I knew what to look out for and how to record the example images in a way that would be consistent for the model.