Few weeks ago, my office had a sharing session. My friend shared how he created a tinder bot to do automatic swiping using selenium. Actually the main point of the presentation is how to use selenium (GUI or script) to finish your daily tasks automatically. But, I was interested with the tinder bot. I though, instead of doing mass swiping to the right or left, we can choose whether he/she is based on our preference or not. If he/she is based on our preference, then swipe right.
After googling some materials, I found some projects about beauty facial prediction, but all projects mostly used SCUT-FBP5500 dataset. The SCUT-FBP5500 dataset has totally 5500 frontal faces with diverse properties, labelled by 60 volunteers. But, what I want to do is to predict the image based on my preference, not by others.
Then, I found this BeautyPredict. This project is quite simple. I tried to understand the process: preparing the data, training, predicting the score, and of course the dataset pattern. After understanding the process, finally I have 2 tasks to be done:
- Create the dataset tool.
- Refactor the BeautyPredict because there are some deprecated function.
Create the dataset tool
I am using flask to build the web app. And, because i am not good at design, thanks to adminlte for helping me with the design. Of course, it is mobile friendly.
And for images data, I have around 3000 face data in my storage, actually I got that data from my research few years ago. Finally, that data has a purpose!
Refactor the beauty predict
Although the project is quite simple, but there are some deprecated functions, so it will be failed if we run it with newer version of the python libraries.
Time to predict
After labeling around 1000 images, I tried to process the images and training the data, then I picked random image from instagram and predict it, it’s quite good and actually the result was as I expected.
So, with this project, we can rate someone whether he/she is based on our preference. If you want to know more about how the system score the image, you can visit the BeautyPredict project on Github.
And I just want to say this on the conclusion, that I believe, all men are handsome and all women are beautiful, but all people have their own preference.
Anyway, this was just my weekend research and it is not finished yet, so I didn’t write the complete instruction. But if you want to know more about the project, you can visit here.