Research map / 3D vision / 3D generation
Pose-ICL: 3D-Aware In-Context Learning for Pose-Controllable Subject Customization
Xuan Han, Yihao Zhao, Mingyu You
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 SAPE instead of concatenated coordinate maps (ablation) | concatenation baseline · 3D-Assets test set | Pose MAE: 22.88 -> 19.33 | Improved |
| add SAPE instead of concatenated coordinate maps (ablation) | concatenation baseline · Real-world test set | Pose MAE: 29.97 -> 23.72 | Improved |
| add SAPE without shallow injection (ablation) | concatenation baseline · 3D-Assets test set | FID: 51.87 -> 60.75 | Hurt |
| add shallow injection (first 19 double-stream layers) (ablation) | SAPE · 3D-Assets test set | Pose MAE: 19.33 -> 14.12 | Improved |
| add shallow injection (first 19 double-stream layers) (ablation) | SAPE · Real-world test set | Pose MAE: 23.72 -> 16.71 | Improved |
| add shallow injection (first 19 double-stream layers) (ablation) | SAPE · 3D-Assets test set | DINO-I: 0.661 -> 0.719 | Improved |
| add sub-patch position embedding (ablation) | SAPE + shallow injection · 3D-Assets test set | Pose MAE: 14.12 -> 16.11 | Hurt |
| add sub-patch position embedding (ablation) | SAPE + shallow injection · Real-world test set | Pose MAE: 16.71 -> 15.78 | Improved |
| add sub-patch position embedding (ablation) | SAPE + shallow injection · Real-world test set | FID: 69.20 -> 56.33 | Improved |
| add balanced segment partitioning (ablation) | SAPE + shallow + sub-patch · 3D-Assets test set | Pose MAE: 16.11 -> 10.75 | Improved |
| add balanced segment partitioning (ablation) | SAPE + shallow + sub-patch · Real-world test set | Pose MAE: 15.78 -> 13.20 | Improved |
| increase reference images from N=4 to 6 or 8 (ablation) | Pose-ICL · Single 3D asset | mean pose error: N=1 average error 13.63 | No clear effect |
Closest papers by meaning
- IC-Custom: Diverse Image Customization via In-Context Learning (ICLR 2026 · similarity 0.87)
- Neural Assets: 3D-Aware Multi-Object Scene Synthesis with Image Diffusion Models (NeurIPS 2024 · similarity 0.85)
- PoseLLaVA: Pose Centric Multimodal LLM for Fine-Grained 3D Pose Manipulation (AAAI 2025 · similarity 0.85)
- ChatPose: Chatting about 3D Human Pose (CVPR 2024 · similarity 0.84)
- UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing (CVPR 2025 · similarity 0.84)
- Disentangled Pose and Appearance Guidance for Multi-Pose Generation (CVPR 2025 · 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.
