When I first got into K-pop, it was only for the music. Then I slowly transitioned into getting to know the members of the groups I liked the most. As someone who is new into K-pop, it is hard getting to know who is who when some groups have over 4 members. And even as someone who has been enjoying the music for over 5 years, I struggle to recognize the differences in groups with over 9 members (especially with the style concepts of each release for singles, eps, or albums).
WJSN has 10 members
Iz*One had 12 members
TripleS has 24 members!
I have always been interested in facial recognition and had prior knowledge of python. This video by Sentdex gave me an amazing starting point and understanding of how to use the face recognition library along with opencv and os.
So far it is only working with NewJeans. Before I add more groups, I want to make it more precise.
I want to add more groups, incorporate scikit-learn to make it more precise, turn it into a full developed web app so others can upload their own images and have those analyzed for recognition.
I am attempting to find ways to optimize face-recognition without using extensive amount of training data. Sometimes, a app will recognize everyone as the same person. I made a hotfix where if a member has already been used to label a face, then that member cannot be used once again. The issue with this approach is that all faces do not get labelled. I am currently working on fixing that too. I'm thinking of comparing and switching based on which face encoding is more accurate.
