Monocular, One-stage, Regression of Multiple 3D People

This paper focuses on the regression of multiple 3D people from a single RGB image. Existing approaches predominantly follow a multi-stage pipeline, which first detects people with the bounding boxes and then regresses their 3D body meshes. In contrast, we propose to Regress all meshes in a One-stage fashion for Multiple 3D People (termed ROMP), which is conceptually simple, bounding box-free, and able to learn per-pixel representation in an end-to-end manner. Our method simultaneously predicts a Body Center heatmap and a Mesh Parameter map, which can jointly describe the 3D body mesh on the pixel level. Through a body-center-guided sampling process, the body mesh parameters of all people in the image can be easily extracted from the Mesh Parameter map. Equipped with such a fine-grained representation, our one-stage framework is free of the complex multi-stage process and more robust to occlusion. Compared with the state-of-the-art methods, ROMP achieves superior performance on the challenging multi-person/occlusion benchmarks, including 3DPW, CMU Panoptic, and 3DOH50K. Experiments on crowded/occluded datasets demonstrate the robustness under various types of occlusion. It is also worth noting that our released demo code ( https://github.com/Arthur151/ROMP ) is the first real-time (over 30 FPS) implementation of monocular multi-person 3D mesh regression to date.