search

hysts / pytorch_mpiigaze_demo

Gaze estimation using MPIIGaze and MPIIFaceGaze
MIT License
290 stars 66 forks source link

different results if recorded video is used #34

Closed jasony93 closed 1 year ago

jasony93 commented 1 year ago

Hello,

I am facing a weird problem. The results I got from real-time inference and the results from recorded video are different. I looked at the same spots from same position and used same camera. The real-time inference shows reasonable results, but the results for the recorded video seems that the gaze has some amount of offset (it follows the gaze but with some offset)

Do you have any idea what the problem is?

jasony93 commented 1 year ago

So far, I figured out that my real-time inference stream and recorded video inference have different resolutions. It turns out that camera matrix has to be modified depending on the resolution. (that's what google said)

hysts commented 1 year ago

Hi, @jasony93 Sorry for the late reply. I just noticed your comments. It seems you have already solved the problem yourself. I think that's the cause too.