Supplementary Material Multiview Human Body Reconstruction from Uncalibrated Cameras

We randomly select samples from them with a ratio of 0.8: 0.2 probabilistically for each batch. We show some qualitative comparisons with another multiview method [15] and a single-view method [9] with post-hoc multiview fusion (average of independent predictions from each view) in Figure 1. Each method reconstruct a unified mesh from multiview inputs. Our method makes better predictions for most cases, which shows the effectiveness of our multiview fusion strategy. We perform additional quantitative evaluations on two challenging in-the-wild multiview dataset, Ski-Pose [20] and MannequinChallenge [14], for comparison with several baseline methods.