Open isarandi opened 1 year ago
I don't know exactly why there is this misalignment. I know the camera parameters were calculated with the checkboard you see in the images, so they should be correctly estimated. Also, the 3D keypoints are consistent in the 3D space. Note that the parameters have been used with Voxelpose to make the annotations, maybe inside the Voxelpose script, there is some preprocessing or postprocessing on the data or on the parameters. You should start investigating from their source code to understand the possible processing that has been performed to understand this error. If you just need the 3D poses for forecasting or action recognition, camera parameters should not be mandatory.
Hi,
Thanks for this work, it's very interesting (I saw your poster at ECCV). I've downloaded the dataset, but the camera calibration seems to be off when I try to visualize poses projected onto the video frames. For example, for the sequence
S00_hammer
, I get these:Do you know of any issue that may cause this problem?