PCN for iOS

[elhoangvu]

Hi guys,
I just tried� porting PCN to iOS and it worked. But it is hard to detect my face as well as the speed is not very good. I use a caffe version for iOS unofficially and only running on the CPU, that may be the reason.
I am ready to welcome everybody to improve it.

And,
Thank you to Jack for your opensource.

---

My repo: https://github.com/elhoangvu/PCN-iOS

[elhoangvu]

I just updated my code from @jack-cv update and it works pretty well with new tracking face. But it is still hard to detect my face.

[Rock-100]

Hi, @elhoangvu, thanks for porting PCN to iOS, it's very cool!

As for the speed problem, firstly, we may need to do many engineering efforts (code optimization, e.g.) in order to make the program run smoothly on mobile phone. I don't know if caffe-ios is efficient enough. Secondly, I guess you can set a larger minFaceSize (65, e.g.), since we don't need to detect small faces on mobile applications normally.

As for the failure cases of detection, could you upload some examples, so I can make further analysis.

Thanks for your work and I welcome everyone to improve it also.

[elhoangvu]

Hi @jack-cv,
First, thank you very much for your enthusiastic support.
The face detection speed issue has been overcome with your new face tracker.
However, the rate of not detection some faces is high, mainly, face has an unstable color when �convert to gray images such as faces that are dazzling, obscured, ... If faces are wearing glasses, the failure rate is quite high.

Again, thank you and your work very much.

This is my test video: https://drive.google.com/file/d/1indGzlcVNQKZVFLUmMlSx60HQhfBp6Q-/view?usp=sharing

[Rock-100]

@elhoangvu Hi, I have seen your video. Indeed PCN has some bad cases, and I think adding some training samples for these cases will be helpful. I will clean up and release the training code when I have time.

[elhoangvu]

OK @jack-cv, thank you very much!