Research map / 3D vision / Human pose estimation
GeoMan: Temporally Consistent Human Geometry Estimation using Image-to-Video Diffusion
Gwanghyun Kim, Xueting Li, Ye Yuan, Koki Nagano, Tianye Li, Jan Kautz, Se Young Chun, Umar Iqbal
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 |
|---|---|---|---|
| use I2G + V2G decomposition instead of naive video diffusion at 30K steps (ablation) | naive extension · validation normals | mean angular error: 19.669 -> 16.185 | Improved |
| use I2G + V2G decomposition with 1/10 training data (ablation) | naive extension · validation normals | mean angular error: 22.329 -> 17.679 | Improved |
| use I2G + V2G decomposition with 1/10 training data (ablation) | naive extension · validation depth | AbsRel: NaN -> 0.013 | Improved |
| train multimodal V2G for normal and depth rather than unimodal (ablation) | unimodal V2G · validation normals | mean angular error: 16.260 -> 16.185 | Improved |
| train multimodal V2G rather than unimodal (ablation) | unimodal V2G · validation normals | TC-Mean: 7.190 -> 6.876 | Improved |
| replace I2G first frame with ground-truth first frame (ablation) | GeoMan I2G+V2G · validation normals | mean angular error: 16.185 -> 14.028 | Improved |
| replace I2G first frame with ground-truth first frame (ablation) | GeoMan I2G+V2G · validation normals | OPW: 0.070 -> 0.078 | Hurt |
| use ground-truth root depth instead of VIMO estimate | GeoMan · ActorsHQ metric depth | delta1: 0.972 -> 0.998 | Improved |
| GeoMan vs Sapiens fine-tuned on same data | Sapiens (fine-tuned) · ActorsHQ normals | mean angular error: 18.338 -> 16.185 | Improved |
| GeoMan vs Sapiens on 30 deg accuracy | Sapiens · ActorsHQ normals | accuracy within 30 deg: 89.177 -> 87.784 | Hurt |
| fine-tune Sapiens on public data instead of proprietary | Sapiens · ActorsHQ normals | mean angular error: 16.278 -> 18.338 | Hurt |
| GeoMan vs Metric3Dv2 for metric depth | Metric3Dv2 · ActorsHQ metric depth | delta1: 0.491 -> 0.972 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Depth Pro — Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025) · better on 1 of 1 shared result
- GeoWizard — GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GeoVideo: Introducing Geometric Regularization into Video Generation Model (NeurIPS 2025 · similarity 0.86)
- GeneMAN: Generalizable Single-Image 3D Human Reconstruction from Multi-Source Human Data (NeurIPS 2025 · similarity 0.86)
- MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (ICLR 2025 · similarity 0.86)
- Reconstructing People, Places, and Cameras (CVPR 2025 · similarity 0.86)
- Seurat: From Moving Points to Depth (CVPR 2025 · similarity 0.86)
- PPMStereo: Pick-and-Play Memory Construction for Consistent Dynamic Stereo Matching (NeurIPS 2025 · 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.
