chingswy
commented
1 year ago Hello, thanks for your attention. In this work, we record the multi-view videos in the wild with unsynchronized cameras. Therefore, the human parameters are not the same among all views.
metro-merl
Hi @chingswy, @cvxzhou, and others, I had a query about the pre-processed soccer data provided as part of the code release of the paper, "Novel View Synthesis of Human Interactions from Sparse Multi-view Videos". In it, the pre-processed smpl json files seem to have been obtained one per camera, this seems slightly counter-intuitive since the SMPL model is a 3D representation of a human and should not be dependent on the particular view, as shown here. Can someone please clarify this?
Appreciate the help!