When passing a video in pipeline the predictions of 3D overlay is fine. But when getting the 2D projected keypoints and mapping that onto the image plane I have noticed a slight displacement in the keypoints being plotted. Is the result from shapy be slightly displaced.
We haven't tested our method with video. However, if you extract the frames, run shape per frame and stack the result you should get the correct output.
When passing a video in pipeline the predictions of 3D overlay is fine. But when getting the 2D projected keypoints and mapping that onto the image plane I have noticed a slight displacement in the keypoints being plotted. Is the result from shapy be slightly displaced.