About label misalignment and projection matrices

[kakekkake]

Hello, sorry for the disturbance. I'm a student studying computer vision. I have some questions related to this dataset. I have implemented this using your code, but for the frames taken by camera 1, the labels that are the correct answers are out of alignment with the events, causing problems with the quality of the posture estimation. Specifically, the labels appear to be shifted parallel to the top of the overall image. Please see the attached image. (These images are subject8_session2_mov2_frame28_cam1 and subject_10_session2_mov2_frame28_cam1) I am considering the possibility that there is a difference in the values of the projection matrix and would appreciate any values that would give us the exact position of the labels. I would also like to know if there is any other way to solve this problem.

Thanks for your reply!

[tobidelbruck]

Thanks for asking about this. The only person who might be able to help is Enrico....
Do you mean that for most frames the alignment appears OK? Could it be that for these frames the subject was moving very quickly and thus perhaps created a misalignment between Vicon and DVS frames?

[kakekkake]

Thank you for replying. Yes, of the data frame consisting of 4 cameras, the labels for cameras 0, 2, and 3 appear to be correct, but only camera 1 appears to be consistently mislabeled. I considered that the cause of the misaligned labels could be due to the subject moving quickly, but since the labels projected to the other cameras in the same scene were fine, I thought the current problem was the projection matrix, not that. Please see the attached image. (These are the same scenes as above, transferred from camera 2) Would you be able to put me in touch with Mr.Enrico?

[enrico-c]

Hi @kakekkake , please take a look at this other issue related to what you're observing https://github.com/SensorsINI/DHP19/issues/11