Research map / 3D vision / Human pose estimation
BEDLAM: A Synthetic Dataset of Bodies Exhibiting Detailed Lifelike Animated Motion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train CLIFF with BEDLAM and fine-tune on 3DPW | CLIFF · 3DPW | MPJPE: 73.9 -> 72.0 | Helped |
| Train CLIFF with BEDLAM and fine-tune on 3DPW | CLIFF · RICH | MPJPE: 90.0 -> 84.5 | Helped |
| Pretrain the backbone on COCO pose estimation rather than ImageNet (ablation) | CLIFF trained on BEDLAM+AGORA · 3DPW | PA-MPJPE: 51.8 -> 47.4 | Helped |
| Pretrain the backbone on COCO pose estimation rather than training from scratch (ablation) | HMR trained on BEDLAM+AGORA · 3DPW | PA-MPJPE: 67.9 -> 47.6 | Helped |
| Increase BEDLAM training crops from 5% to 100% (ablation) | CLIFF · 3DPW | PA-MPJPE: 54.0 -> 50.5 | Helped |
| Increase BEDLAM training crops from 5% to 100% (ablation) | HMR · 3DPW | PA-MPJPE: 55.8 -> 51.2 | Helped |
| Train on BEDLAM rather than AGORA alone (ablation) | CLIFF · 3DPW | PA-MPJPE: 54.0 -> 50.5 | Helped |
| Train on BEDLAM rather than AGORA alone (ablation) | HMR · 3DPW | PA-MPJPE: 58.3 -> 51.2 | Helped |
| Combine BEDLAM and AGORA training data instead of using BEDLAM alone (ablation) | CLIFF · 3DPW | PA-MPJPE: 50.5 -> 47.4 | Helped |
| Combine BEDLAM and AGORA training data instead of using BEDLAM alone (ablation) | HMR · 3DPW | PA-MPJPE: 51.2 -> 47.6 | Helped |
| Use realistic simulated clothing rather than textured bodies (ablation) | HPS regressors trained on synthetic data · 3DPW | accuracy | Helped |
| Replace the CLIFF architecture with basic HMR while keeping BEDLAM training | HPS regression · RICH | MPJPE: 90.0 -> 91.4 | Mixed |
Closest papers by meaning
- BEDLAM2.0: Synthetic humans and cameras in motion (NeurIPS 2025 · similarity 0.94)
- SyncHuman: Synchronizing 2D and 3D Generative Models for Single-view Human Reconstruction (NeurIPS 2025 · similarity 0.87)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.86)
- 4D-DRESS: A 4D Dataset of Real-World Human Clothing With Semantic Annotations (CVPR 2024 · similarity 0.86)
- PKU-DyMVHumans: A Multi-View Video Benchmark for High-Fidelity Dynamic Human Modeling (CVPR 2024 · similarity 0.86)
- SCULPT: Shape-Conditioned Unpaired Learning of Pose-dependent Clothed and Textured Human Meshes (CVPR 2024 · similarity 0.86)
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.