Relative Pose Estimation for RGB-D Human Input Scans via Implicit Function Reconstruction

Figure 3

We show the performance of our multi-view extension. The first two rows show the 3-view input scans while the last two rows with 4-view input scans. We still add different small disturbances to the groundtruth poses. The first column is the reconstruction with original pose, the second column is the reconstruction with only pairwise optimization, the third column is the human reconstruction with our proposed optimization method, and the last column with the groundtruth pose.