HMOR: Hierarchical Multi-person Ordinal Relations for Monocular Multi-person 3D Pose Estimation

Putting People in their Place: Monocular Regression of 3D People in Depth

“…We evaluate BEV on two multi-person datasets: inthe-wild using the 2D RH and in 3D using the synthetic AGORA [24] dataset. On RH, compared with previous methods [10,21,34,41], BEV is more accurate in relative depth reasoning and pose estimation. On AGORA, BEV significantly improves detection and achieves state-of-theart results on "AGORA kids" in terms of the mesh reconstruction error.…”

“…While previous multi-person methods perform well in constrained experimental settings, they struggle with severe occlusion, diverse body size and appearance, the ambiguity of monocular depth, and in-the-wild cases [10,21,34,39,41]. These challenges lead to unsatisfactory performance in crowded scenes, including detection misses, similar predictions for overlapping people, and all predictions having a similar height.…”

“…However, current methods lack sufficiently powerful representations to learn from these cases. A few learning-based methods have been proposed for depth reasoning of predicted body meshes [10,39] or 3D poses [21,34,41]. Unfortunately, they all reason about depth via 2D representations, such as RoI-aligned features [10,21], a 2D depth map [34,41], or multi-scale 2D center maps [39].…”

“…A few learning-based methods have been proposed for depth reasoning of predicted body meshes [10,39] or 3D poses [21,34,41]. Unfortunately, they all reason about depth via 2D representations, such as RoI-aligned features [10,21], a 2D depth map [34,41], or multi-scale 2D center maps [39]. These regression-based 2D representations have inherent drawbacks for representing the 3D world.…”

Shape-aware Multi-Person Pose Estimation from Multi-View Images

“…When only one camera is available, the problem is underdetermined since many 3D poses may correspond to the same 2D pose. Leveraging the learning-based method, 3D poses can be recovered by lifting detected 2D poses [41,42,55], or directly regressing 3D poses [3,13,48,53], or by fitting parametric human body models [21,53]. However, the reconstruction accuracy of these methods is limited due to the depth ambiguities and strong occlusions when multiple humans are close to each other.…”