DGG-HMR: Multi-Person Human Mesh Recovery with Depth-Guided Geometric Anchoring

Multi-person human mesh recovery (HMR) from a single image is inherently ill-posed, as multiple 3D poses can produce identical 2D projections due to depth ambiguity. Most existing methods implicitly regress 3D translation from image features, which often leads to unreliable depth estimation. To address this issue, we propose a depth-guided multi-person HMR framework that explicitly models instance-level depth cues and integrates them into mesh recovery. Specifically, we first introduce an instance-aware depth estimator to predict per-person pelvis depths that serve as explicit 3D anchors, thereby decoupling depth estimation from mesh regression. Then, we design a geometry-anchored refinement decoder that uses these anchors to initialize each instance within a plausible 3D neighborhood, stabilizing mesh refinement under joint 2D-3D supervision. Finally, we adopt a single-stage joint training strategy to coordinate depth estimation and mesh recovery in a unified framework. Extensive experiments on multiple benchmarks demonstrate that our method achieves state-of-the-art performance in both mesh reconstruction accuracy and depth ordering.

This paper studies how to recover the 3D body shapes and positions of multiple people from a single image. This is difficult because a 2D image does not directly reveal how far each person is from the camera. Different 3D body arrangements can look almost the same in the image, especially in crowded scenes, which can cause existing methods to place people at incorrect depths or produce unstable spatial layouts. We propose DGG-HMR, a method that explicitly estimates the depth of each person and uses this information to guide 3D human reconstruction. Instead of relying only on image appearance, our method first predicts where each person is and how far their body center is from the camera. These depth cues are then used as geometric anchors to help reconstruct each person in a more plausible 3D location. This makes the reconstruction process more stable and improves the relative depth ordering among people. Experiments on several standard benchmarks show that our method improves both 3D body reconstruction accuracy and spatial ordering in multi-person scenes. The results suggest that explicitly modeling per-person depth is an effective way to reduce ambiguity in single-image 3D human understanding.