Research map / 3D vision / Human pose estimation
Humans as a Calibration Pattern: Dynamic 3D Scene Reconstruction from Unsynchronized and Uncalibrated Videos
Changwoon Choi, Jeongjun Kim, Geonho Cha, Minkwan Kim, Dongyoon Wee, Young Min Kim
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 |
|---|---|---|---|
| initialize camera poses and time offsets from human motion alignment, with progressive joint optimization, instead of Sync-NeRF with GT poses | dynamic K-Planes reconstruction · Panoptic Studio average | PSNR: 21.19 -> 26.55 | Improved |
| add progressive learning strategy (coarse-to-fine weighting and curriculum of parameters) (ablation) | joint NeRF and calibration optimization · Panoptic Studio average | PSNR: 19.66 -> 26.55 | Improved |
| add progressive learning strategy (ablation) | joint NeRF and calibration optimization · Mobile-Stage DANCE | PSNR: 13.72 -> 23.27 | Improved |
| refine calibration jointly with dynamic NeRF after human-motion initialization | calibration pipeline · Panoptic Studio average | rotation error (deg): 5.46 -> 0.412 | Improved |
| refine time offsets with joint optimization | calibration pipeline · Panoptic Studio average | time offset error (frames): 1.343 -> 0.028 | Improved |
| use L2 distance of 3D joint positions instead of L2 distance of SMPL parameters for time offset matching (ablation) | time offset initialization · Panoptic Studio TENNIS | time offset error (frames): 2.100 -> 0.467 | Improved |
| add Gaussian noise to SMPL human motion (sigma up to 0.05) (ablation) | initialization stage · Panoptic Studio TENNIS | time offset error (frames): 0.467 -> 0.833 (sigma 0.05) | No clear effect |
| add severe SMPL noise (sigma 0.1 and 0.2) (ablation) | initialization stage · Panoptic Studio TENNIS | time offset error (frames): 0.467 -> 2.100 (0.1), 2.367 (0.2) | Hurt |
| degrade input videos or use mixed frame rates (ablation) | initialization stage · Panoptic Studio TENNIS | translation error (cm): 25.827 -> 32.374 (degraded), 38.58 (mixed FPS) | Hurt |
| joint optimization with time offsets instead of GT calibration (compared with oracle) | proposed method · Mobile-Stage DANCE | PSNR: 22.98 -> 23.27 | No clear effect |
Closest papers by meaning
- Learning Neural Volumetric Representations of Dynamic Humans in Minutes (CVPR 2023 · similarity 0.90)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.89)
- PKU-DyMVHumans: A Multi-View Video Benchmark for High-Fidelity Dynamic Human Modeling (CVPR 2024 · similarity 0.89)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.89)
- MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild (CVPR 2024 · similarity 0.88)
- Reconstructing People, Places, and Cameras (CVPR 2025 · similarity 0.88)
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.
