Hello,
Disclaimer: This is not about identiy finding - we dicussed that very long and successfull("Facenet", "Siamese Networks") here on open cv.
I am currently importing and processing(creating embeddings) images of popular persons. Some of these popular persons i am interested in are not the only person in the image(some crowd for a wrestler image for example) and my face detector gets multiple detections.
Its a bit sad - i dont want to skip these pictures. Hmm what can i do to get the right? Some trivial thoughts:
- pick biggest bounding box
- pick bounding box with highes confidence
I could also combine these two things and maybe thats already enough? Any thoughts / hints on this?
Thank you a lot + Greetings, Holger