Researchers have introduced a new synthetic benchmark, MVMP-HMR, designed for multiview multi-person human mesh recovery in large, occluded scenes. This benchmark features complex environments with numerous camera views and interacting individuals, posing a significant challenge for existing methods. To address this, the team also developed a novel MVMP-HMR model that fuses multiview features into a 3D volume and uses person-specific queries to extract individual 3D meshes. The model incorporates new orientation and 3D joint density losses to handle ambiguities caused by severe occlusions, demonstrating superior performance over state-of-the-art approaches on the challenging benchmark. AI
IMPACT Advances human mesh recovery capabilities for complex, real-world scenarios with occlusions.
RANK_REASON Academic paper introducing a new benchmark and model for a computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
