41 points by tensor_face 6 months ago flag hide 12 comments
johnreed 6 months ago next
Great work! The real-time face recognition system is very impressive. I'm curious, what kind of hardware are you using to achieve real-time results?
philipbranch 6 months ago next
Hi @johnreed, thanks for your comments! I'm using a NVIDIA RTX 3090 for this project. It helps a lot with the real-time experience.
alexcloud 6 months ago prev next
This is impressive! I'd also like to know how well it can detect faces at different angles and lighting conditions.
franktensor 6 months ago next
Hey @alexcloud, I tested it in various light conditions and collected some images under harsh light as well as in low light. The accuracy is surprisingly good! You can find more details in the repository.
chanhacker 6 months ago prev next
But is it GPU-accelerated? I know TensorFlow 2.0 supports GPU acceleration.
philipbranch 6 months ago next
@chanhacker, yes, indeed, this project fully leverages the benefits of GPU acceleration. It greatly helps in real-time video use cases.
marielearner 6 months ago prev next
Did you create a user-friendly interface for your project? It seems like this system could have multiple use cases.
philipbranch 6 months ago next
Hey @marie learner, glad you think so! At this stage, the system is command-line based, but I plan to create a web interface and even mobile apps as extensions of the project.
jaesungdude 6 months ago prev next
Have you considered using other deep learning frameworks for a comparison? PyTorch might also do well with face recognition.
philipbranch 6 months ago next
True, @jaesungdude, I tried a simple POC with PyTorch, and it actually performed on parity with TensorFlow, considering my personal implementation. However, TensorFlow is more mature and better documented for this type of use case.
rachel2code 6 months ago prev next
Impressive demo! I cannot wait to see what you will implement in future updates. This could help improve various video content moderation systems.
philipbranch 6 months ago next
Thanks for your suggestion, @rachel2code. I'm excited about the potential too, and it's great to hear that people see its potential for applications in this field.