How do you know which selfie photo to post in Instagram if you took 10-20 photos at the same time, but with different emotions and poses? In this case, the HowCuteAmI application will help. Based on neural networks, it is able to determine how beautiful you are. Don't believe me? Then we are ready to describe how the application works, based on the material posted by the developer himself.
This application will help you understand how beautiful you are.
The project outline looks like this:
The developer – Alexander Kharin – used the Python language to train the neural network and a sample with marked-up photographs of people, where their beauty is assessed, was collected in one of the universities in China. There are not so many photographs in the sample – 5500 pieces. Of these, only 750 are Caucasoid faces, which is very few, but already now the application is showing good results.
According to a standard deviation of 0.64, the author of the material came to the conclusion that appraisers from a Chinese university almost always gave similar scores to different photos, which suggests that beauty is not individual and for all people the beauty parameters are about the same, so you can approach this question mathematically.
Next, he developed a neural network of 32 neurons that converts a vector of image features into a final score describing a person's beauty. He trained her on the basis of data from the University of China and got a working neuron. Alexander developed a small Android – HowСuteAmI application, so that you can quickly evaluate the application, and we have already tested it.
I would like to say right away that while the application, at least in my case, does not work very well. The front camera of my device produces images of 10-20 megabytes in size, and the application crashes if I try to upload such a large photo, so I have to compress it.
What's the bottom line? I tried a few of my own images and photographs of various people. The results are interesting, but not without questions. As for my photos, the application gave them a rather accurate rating in accordance with how I rated them myself, because a lot depends on the lighting and angle, and in this matter the application can accurately predict which photo from a sample of photos of one person will be the best. . The maximum score I managed to squeeze is 7.5, on average my photos are gaining 5-7 points. At the same time, I honestly do not consider myself a handsome man, and it is very strange that, for example, Johnny Depp received such a low score, given that objectively he is a handsome person.
The same goes for Angelina Jolie. Of course she is beautiful, and I would give her 9-10 points, but the application thinks differently. Probably the reason is the small sample, 750 images of Caucasoid faces are not enough for the application to accurately determine the beauty of a person.
The result is still difficult to call final, but such a study says only one thing – the tastes of people in the world are similar, and it is possible that in the future, based on likes in photos Facebook, it will be able to develop its own neural network and embed it into Instagram. In this case, before posting, users will be able to easily assess how successful the selfie turned out to be. And the quality of the assessment will be as objective as possible. It's great when there are such applications and such developments. They make the world a better place and, most importantly, more beautiful.
It is interesting to know the opinion of our readers. How many points did you get? Share your results in the comments to this publication or in our Telegram chat.
- Application: HowСuteAmI
- Developer: Alexander Kharin
- Category: Entertainment
- Version: 1.0.10
- Price: Free
- Download: Google Play