Research map / 3D vision / Human pose estimation
The One Where They Reconstructed 3D Humans and Environments in TV Shows
Georgios Pavlakos, Ethan Weber, Matthew Tancik, Angjoo Kanazawa
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 intrinsics and extrinsics to multi-shot optimization (calibrated vs uncalibrated) (ablation) | multi-shot optimization · Human3.6M | MPJPE (mm): 131.9 -> 65.8 | Improved |
| add extrinsics on top of intrinsics in multi-shot optimization (ablation) | multi-shot optimization · Human3.6M | MPJPE (mm): 123.8 -> 65.8 | Improved |
| add camera calibration to multi-shot optimization (ablation) | multi-shot optimization · TV shows | pelvis distance error (cm): 889cm -> 38cm | Improved |
| add extrinsics on top of intrinsics in multi-shot optimization (ablation) | multi-shot optimization · TV shows | pelvis distance error (cm): 59cm -> 38cm | Improved |
| add camera intrinsics to contextual monocular fitting (ablation) | SMPLify with context · TV shows | cross-shot PCK (%): 16.0% -> 88.7% | Improved |
| add body shape scale from neighboring shot change (ablation) | SMPLify with context · TV shows | cross-shot PCK (%): 65.9% -> 88.7% | Improved |
| add NeRF density structure term to avoid interpenetration (ablation) | SMPLify with context · TV shows | cross-shot PCK (%): 87.5% -> 88.7% | Improved |
| add full 3D context to SMPLify | SMPLify · TV shows | cross-shot PCK (%): 16.5% -> 88.7% | Improved |
| replace image-based re-ID costs with multi-shot fitting cost incl. anthropometric constraints | re-ID matching · TV shows shot-boundary re-ID | Re-ID F1: 0.80 -> 0.91 | Improved |
| add anthropometric constraints to keypoint triangulation cost (ablation) | re-ID matching · TV shows shot-boundary re-ID | Re-ID F1: 0.86 -> 0.91 | Improved |
| use 3D human reconstruction for gaze instead of face-based gaze following | gaze estimation · TV shows gaze following (all) | PCGD (%): 16% -> 62% | Improved |
| use 3D human reconstruction for gaze instead of face-based gaze following | gaze estimation · TV shows gaze following (w/ face) | PCGD (%): 32% -> 67% | Improved |
Closest papers by meaning
- ShowMak3r: Compositional TV Show Reconstruction (CVPR 2025 · similarity 0.87)
- Reconstructing People, Places, and Cameras (CVPR 2025 · similarity 0.84)
- Guess The Unseen: Dynamic 3D Scene Reconstruction from Partial 2D Glimpses (CVPR 2024 · similarity 0.83)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.83)
- MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures (CVPR 2024 · similarity 0.83)
- From an Image to a Scene: Learning to Imagine the World from a Million 360° Videos (NeurIPS 2024 · similarity 0.83)
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.
