Research map / 3D vision / Face and shape
PersonaCraft: Personalized and Controllable Full-Body Multi-Human Scene Generation Using Occlusion-Aware 3D-Conditioned Diffusion
Gwanghyun Kim, Suh Yoon Jeon, Seunggyu Lee, Se Young Chun
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 depth SMPLx conditioning instead of RGB (ablation) | SCNet · Pose ablation | MPJPE (3D): 82.52 -> 71.20 | Improved |
| use depth SMPLx conditioning instead of normal (ablation) | SCNet · Pose ablation | MPJPE (3D): 79.30 -> 71.20 | Improved |
| combine depth and normal SCNet residuals (ablation) | SCNet-D · Pose ablation | MPJPE (3D): 71.20 -> 63.16 | Improved |
| combine depth and normal SCNet residuals (ablation) | SCNet-D · Pose ablation | AP-0.5 (2D): 0.423 -> 0.494 | Improved |
| add OccNet (ablation) | SCNet-D + SCNet-N · Pose ablation | MPJPE (3D): 63.16 -> 62.92 | Improved |
| add OccCFG (ablation) | SCNet-D + SCNet-N + OccNet · Pose ablation | MPJPE (3D): 62.92 -> 62.64 | Improved |
| add OccCFG (ablation) | SCNet-D + SCNet-N + OccNet · Pose ablation | AP-0.5 (2D): 0.499 -> 0.495 | Hurt |
| use SMPLx body shape parameters (ablation) | PersonaCraft · Body shape preservation (total) | cosine similarity: 0.539 -> 0.615 | Improved |
| replace OMG-based InstantID with PersonaCraft | InstantID + OMG · Multi-human personalization | Face ID preservation (total): 0.227 -> 0.317 | Improved |
| replace 2D pose ControlNet-SDXL with 3D SMPLx conditioning | ControlNet-SDXL · Pose-controlled generation | MPJPE (3D): 114.93 -> 62.647 | Improved |
| fine-tune ControlNet-SDXL baseline on MPII data (ablation) | ControlNet-SDXL · Pose-controlled generation | MPJPE (3D): 114.93 -> 100.817 | Improved |
| replace ControlNet-SDXL with PersonaCraft on text alignment | ControlNet-SDXL · Pose-controlled generation | CLIP sim: 0.285 -> 0.274 | Hurt |
Closest papers by meaning
- Visual Persona: Foundation Model for Full-Body Human Customization (CVPR 2025 · similarity 0.89)
- PSHuman: Photorealistic Single-image 3D Human Reconstruction using Cross-Scale Multiview Diffusion and Explicit Remeshing (CVPR 2025 · similarity 0.87)
- PersonaHOI: Effortlessly Improving Face Personalization in Human-Object Interaction Generation (CVPR 2025 · similarity 0.86)
- OccFusion: Rendering Occluded Humans with Generative Diffusion Priors (NeurIPS 2024 · similarity 0.85)
- FRESA: Feedforward Reconstruction of Personalized Skinned Avatars from Few Images (CVPR 2025 · similarity 0.84)
- IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation (NeurIPS 2024 · similarity 0.84)
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.
