Research map / 3D vision / Face and shape
Direct Reward Fine-Tuning on Poses for Single Image to 3D Human in the Wild
Seunguk Do, Minwoo Huh, Joonghyuk Shin, Jaesik Park
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 |
|---|---|---|---|
| post-train PSHuman with DrPose on DrPose15K | PSHuman · MixamoRP | Chamfer Distance: 137.2814 -> 126.5312 | Improved |
| post-train PSHuman with DrPose on DrPose15K | PSHuman · THuman2.1-test | Chamfer Distance: 52.9643 -> 42.0529 | Improved |
| post-train PSHuman with DrPose on DrPose15K | PSHuman · CustomHumans-test | Chamfer Distance: 52.2187 -> 44.1326 | Improved |
| post-train PSHuman with DrPose on DrPose15K | PSHuman · THuman2.1-test | PSNR: 18.3880 -> 20.8594 | Improved |
| post-train PSHuman with DrPose on DrPose15K | PSHuman · MixamoRP | Normal Consistency: 0.5876 -> 0.5998 | Improved |
| post-train Era3D* with DrPose on DrPose15K | Era3D* · MixamoRP | Chamfer Distance: 150.0118 -> 126.0622 | Improved |
| post-train Era3D* with DrPose on DrPose15K | Era3D* · THuman2.1-test | PSNR: 17.7662 -> 20.3563 | Improved |
| use PSHuman instead of Era3D as base model for DrPose (ablation) | DrPose · THuman2.1-test / CustomHumans-test / MixamoRP | CD, NC, PSNR: CD 41.1770 vs 42.0529 (THuman2.1) | No clear effect |
| add DrPose15K pose data (1.73x larger joint std than THuman2.1) (ablation) | I2MV diffusion · pose diversity | standard deviation of joint positions: 1.73x larger std vs THuman2.1 | Improved |
| add KL divergence regularization to reward fine-tuning (ablation) | DrPose · reward fine-tuning | image quality / reward hacking | 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.
- PSHuman — PSHuman: Photorealistic Single-image 3D Human Reconstruction using Cross-Scale Multiview Diffusion and Explicit Remeshing (CVPR 2025) · better on 7 of 8 shared results
- SiTH — SiTH: Single-view Textured Human Reconstruction with Image-Conditioned Diffusion (CVPR 2024) · better on 3 of 3 shared results
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
- MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild (CVPR 2024 · similarity 0.87)
- GeneMAN: Generalizable Single-Image 3D Human Reconstruction from Multi-Source Human Data (NeurIPS 2025 · similarity 0.87)
- FinePOSE: Fine-Grained Prompt-Driven 3D Human Pose Estimation via Diffusion Models (CVPR 2024 · similarity 0.87)
- PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 2025 · similarity 0.87)
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.86)
- COMPOSE: Hypergraph Cover Optimization for Multi-view 3D Human Pose Estimation (NeurIPS 2026 · 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.
