Interesting to compare this against VideoPose3D, which claims half the MPJPE that this paper does. I believe the difference is that this method can only examine a single frame, whereas the other method looks at multiple frames.
I dislike the way they cut off their graphs, here’s the MPJPE with 0 included in the Y-axis:
Interesting to compare this against VideoPose3D, which claims half the MPJPE that this paper does. I believe the difference is that this method can only examine a single frame, whereas the other method looks at multiple frames.
I dislike the way they cut off their graphs, here’s the MPJPE with 0 included in the Y-axis: