search

arnaldog12 / Deep-Learning

Study and implementation about deep learning models, architectures, applications and frameworks
83 stars 43 forks source link

Question: Re-training using 5 landmark points #7

Closed santo4ul closed 4 years ago

santo4ul commented 4 years ago

Hi, Thanks for the inputs you shared on #6. The result looks good. Training was quite easy!

I'm planning to use the same method, but only with 5 landmark points (left and right eye center, nose tip, right and left mouth corners). The reason is that, MTCNN face detector outputs 5 landmark points and I'm planning to re-use the same and there by avoiding the dlib shape predictor.

Could you share your inputs? I agree that the number of features with 5 points would be very less when compared to 68 points. However, I'd like to get your inputs.

Thanks, San

arnaldog12 commented 4 years ago

I'm sorry, @santo4ul, but I don't have the inputs anymore neither the code to generate them. However, it's easy to use Dlib in Python to detect faces. You can use either the this post or this repo as reference.

You can also check my Medium post to see more details about how I computed the features.

Please, close this issue if you have no more questions.

santo4ul commented 4 years ago

Thanks a lot @arnaldog12 for your inputs. I'll close the ticket.