Closed yuyaya-foifoi closed 3 years ago
@yuyaya-foifoi
I thought it was written in the paper, but it seems not. I can't remember where it came from. Hmm... Maybe from some code?
@hysts
Thank you for your reply. I see, do you remember which code you referred to ?
@yuyaya-foifoi
I don't remember, but anyway the code on the project page seems to use different parameters, so I think the parameters in this repo are incorrect.
@hysts
It's a parameter in the HeadPoseEstimator in headposeestimator.cpp, right?
I changed current parameter to the parameter in HeadPoseEstimator and ran demo.py, but qualitatively, the accuracy decreased. (the gaze vector in the image didnt not move.) I also changed parame cx and cy to the size of my camera image, but it was still qualitatively inaccurate.
For now, it seems better not to change the current implementation.
@yuyaya-foifoi
Oh, thanks for the info. As you said, the parameters they are using doesn't seem to be working. I didn't notice it due to a bug in the demo program I used when I wrote the comment above. Sorry about that.
And thanks to you, I remember that the parameters in the project page didn't work when I was developing this repo, and that I googled for a code with parameters that works. And I think this is where the parameter in this repo came from.
@hysts
Thank you for sharing. I see, I understand that the effectiveness of this parameter seems to be empirically proven. If you find out what this parameter theoretically means, please share it. I'll share it if I find out something.
Hi, @hysts thank you again for sharing your work
Can you tell me how you set normalized camera projection matrix?