“Hello Beautiful”, said AI

Introduction

I was searching for some important things on Google when I saw a data set of “Beautiful and Average Women’s Faces“. It made me curious. So, I began to explore the data set, rationally, and got an idea to train a model of Artificial Intelligence, Machine Learning and see what it will predict.

I will not explain all the coding of Machine Learning here as it is not a tutorial but yes, I will tell you people some basics. To keep it simple, I will also avoid jargon. In this blog, I will use the term “machine” for my laptop. In the later part, I will also discuss some limitations of the model.

So let’s start…

Training Machine Learning Model

The data set I downloaded of women’s faces, contained 4000 pictures of faces of various Western women. It contained 2000 pictures of beautiful and 2000 pictures of average-looking women’s faces. I fetched all the pictures in Python and converted those images into numerical form. I also transformed all the pictures into grayscale (Black and white) as the ML model has nothing to do with colors, particularly in this case, but rather with shapes and curves. For better computation, I divided all the numbers by 255. Then I labeled the images so that the machine could classify between both types of women’s faces, beautiful and average-looking. This was a bit difficult for two reasons, technical and non-technical. The technical reason was that the total of 4000 pictures was quite large and required high processing speed. The non-technical reason was, perhaps, that it was data of beautiful ladies. I suppose my “Modest Machine” was a bit reluctant to process images of attractive females. That’s why it kept giving me errors, initially.

Testing The Model

My machine took 3.5 hours to learn all the facial features of beautiful and average ladies. During this duration, I wrote some parts of this blog that you are reading. Now, it was time to validate my ML model. I couldn’t test the model on my image as it was trained on female faces.

Jillz Guerin is one of my favorite YouTubers because of her thoughts, wisdom, femininity, and beauty. She is a traditional woman. She teaches chicks how to be good girls. I tested my ML model on one of her videos.

Thank God, I and AI were on the same page as far as the magnificence of this lady is concerned.

Non-Western Women’s Face Test

I was delighted when my model predicted that Mahira Khan is beautiful. As I mentioned earlier, the model was trained on Western women’s faces. So, it was quite impressive.

After finding Mahira Khan attractive, I decided to test the model on darker skin tones. Here, I found mixed results. One thing should be clear that ML model is not racist regarding beauty standards but it was merely trained on white women’s faces.

Final Remarks

There are two separate things, the accuracy of the model and the nature of the data set. My model was quite accurate because 95% of the time, it recognized faces. Whereas, despite 4000 images in the data set, there was not a clear set of boundaries between beautiful and average-looking facial features. In the folder of the average women’s face training set, many faces were also quite attractive. Therefore, my ML model labeled most of the faces “Beautiful”. I spent many hours coding this ML model. I was eager to watch my model predict anyone “Average”. So, here I found my desired results.

EID MUBARAK!