Research map / 3D vision / Human pose estimation
Multi-RoI Human Mesh Recovery with Camera Consistency and Contrastive Losses
Yongwei Nie, Changzhen Liu, Chengjiang Long, Qing Zhang, Guiqing Li, Hongmin Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add camera consistency loss (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 83.1 -> 80.8 | Improved |
| add contrastive learning module (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 83.4 -> 80.8 | Improved |
| add crop-aware fusion module (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 83.8 -> 80.8 | Improved |
| use multi-crop input without any of the three components (ablation) | CLIFF · 3DPW (trained on COCO) | MPJPE: 84.6 (1 RoI) -> 87.0 | No clear effect |
| use relative positional encoding of bounding boxes for fusion weights (ablation) | Crop-aware fusion · 3DPW (trained on COCO) | MPJPE: 98.7 -> 80.8 | Improved |
| use relative PE instead of absolute bbox PE (ablation) | Crop-aware fusion · 3DPW (trained on COCO) | MPJPE: 87.6 -> 80.8 | Improved |
| increase the number of RoIs from 1 to 5 (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 84.6 -> 80.8 | Improved |
| increase the number of RoIs from 1 to 5 (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | inference speed (fps): 76.3 -> 55.6 | Hurt |
| use fixed shifted and scaled crops instead of random crops (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 84.4 -> 80.8 | Improved |
| add both scaling and shifting to the fixed crops (ablation) | Multi-RoI HMR · 3DPW (trained on COCO) | MPJPE: 83.6 -> 80.8 | Improved |
| replace CLIFF single-RoI with multi-RoI model (HRNet-W48) | CLIFF · 3DPW test | MPJPE: 69.0 -> 64.1 | Improved |
| feed contrastive projection z instead of encoder feature h to later modules (ablation) | Contrastive module · 3DPW | accuracy | Hurt |
Closest papers by meaning
- Monocular Models are Strong Learners for Multi-View Human Mesh Recovery (ECCV 2026 · similarity 0.91)
- End-to-End Recovery of Human Shape and Pose (CVPR 2018 · similarity 0.89)
- FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery (NeurIPS 2026 · similarity 0.89)
- Incorporating Test-Time Optimization into Training with Dual Networks for Human Mesh Recovery (NeurIPS 2024 · similarity 0.88)
- ADHMR: Aligning Diffusion-based Human Mesh Recovery via Direct Preference Optimization (ICML 2025 · similarity 0.88)
- BLADE: Single-view Body Mesh Estimation through Accurate Depth Estimation (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
