Research map / 3D vision / Human pose estimation
MetaFuse: A Pre-trained Fusion Model for Human Pose Estimation
Rongchang Xie, Chunyu Wang, Yizhou Wang
Check 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 |
|---|---|---|---|
| Use MetaFuse instead of NaiveFuse with 50 adaptation images | Cross-view pose fusion · Panoptic | MPJPE: $43.39$mm -> $37.27$mm | Improved |
| Use MetaFuse instead of NaiveFuse with 200 adaptation images | Cross-view pose fusion · Panoptic | MPJPE: $35.60$mm -> $31.78$mm | Improved |
| Use MetaFuse instead of NaiveFuse with 50 adaptation images | Cross-view pose fusion · Panoptic | JDR (Avg): 85.1 -> 88.9 | Improved |
| Use MetaFuse instead of No-Fusion | Cross-view pose fusion · Total Capture | MPJPE (Mean): 41.2 -> 32.4 | Improved |
| Add multi-view fusion with 50 labeled examples | H36M 2D pose estimation · H36M | JDR | Improved |
| Increase MetaFuse adaptation data from 50 to 500 images | H36M 3D pose estimation · H36M | MPJPE: 32.7mm -> 29.3mm | Improved |
| Add temporal information through LSTM-AE instead of MetaFuse | 3D pose estimation · Total Capture Walking2 | MPJPE: 21.7 -> 13.0 | Improved |
| Add Savitzky-Golay smoothing to MetaFuse 3D poses (ablation) | MetaFuse · Total Capture Walking | MPJPE: about 5mm | Improved |
| Use regular SGD pretraining instead of meta-learning for the factorized fusion model (ablation) | AffineFuse · H36M | JDR | Hurt |
| Replace geometry-inspired factorization with a NaiveFuse model, while retaining meta-learning (ablation) | MetaFuse · H36M | JDR (average): 87.7% and 89.3% | Hurt |
| Use NaiveFuse with only 50 target images rather than MetaFuse (ablation) | Cross-view pose fusion · H36M | JDR | Hurt |
| Add L2, L1, or L2,1 regularization to NaiveFuse (ablation) | NaiveFuse · H36M | JDR | No clear effect |
Closest papers by meaning
- AdaFuse: Adaptive Multiview Fusion for Accurate Human Pose Estimation in the Wild (IJCV 2020 · similarity 0.92)
- FusionFormer: A Concise Unified Feature Fusion Transformer for 3D Pose Estimation (AAAI 2024 · similarity 0.87)
- Fast Adaptation for Human Pose Estimation via Meta-Optimization (CVPR 2024 · similarity 0.86)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.84)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.84)
- PoseIRM: Enhance 3D Human Pose Estimation on Unseen Camera Settings via Invariant Risk Minimization (CVPR 2024 · similarity 0.84)
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.
