Research map / 3D vision / 3D generation
PWM-ArtGen: Part World Model for Articulated Object Generation
Wentao Zheng, Ancong Wu
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 image branch jointly modeling visual dynamics (ablation) | PWM-ArtGen · ACD (GT graph) | RS-d_gIoU: 0.6567 -> 0.5948 | Improved |
| add visual dynamics regularizer (VDR) to image branch (ablation) | PWM-ArtGen · ACD (GT graph) | RS-d_gIoU: 0.5948 -> 0.5885 | Improved |
| add co-training on PM-R to image branch (ablation) | PWM-ArtGen · ACD (GT graph) | RS-d_gIoU: 0.5948 -> 0.5786 | Improved |
| add VDR and co-training together (ablation) | PWM-ArtGen · ACD (GT graph) | RS-d_gIoU: 0.5948 -> 0.5693 | Improved |
| add VDR (ablation) | PWM-ArtGen · PM visual dynamics | PSNR: 21.964 -> 23.383 | Improved |
| add co-training (ablation) | PWM-ArtGen · PM visual dynamics | PSNR: 21.964 -> 22.908 | Improved |
| add co-training to IB+VDR (collision metric) (ablation) | PWM-ArtGen · ACD (GT graph) | AOR: 0.0038 -> 0.0031 (VDR), but IB+Co-T 0.0020 best | Mixed |
| PWM-ArtGen vs baselines without GT graph on ACD | PWM-ArtGen · ACD predicted graph | RS-d_gIoU: 0.7574 -> 0.6822 (vs Articulate-Anything) | Improved |
| PWM-ArtGen vs SINGAPO on PM without GT graph | PWM-ArtGen · PartNet-Mobility predicted graph | RS-d_CD: 0.0199 (SINGAPO) vs 0.0446 (ours) | Hurt |
| PWM-ArtGen dynamics branch vs DragAPart | PWM-ArtGen · PM visual dynamics | PSNR/SSIM/LPIPS: PSNR 24.803 -> 24.328; time 6.9 s -> 0.7 s | Mixed |
| drop movable parts from GT graph (d=0 to d=4) (ablation) | PWM-ArtGen · PM | RS-d_cDist: 0.0455 -> 0.1541 | Hurt |
| vary co-training weight (ablation) | PWM-ArtGen · ACD | d_gIoU | No clear effect |
Closest papers by meaning
- SINGAPO: Single Image Controlled Generation of Articulated Parts in Objects (ICLR 2025 · similarity 0.91)
- DIPO: Dual-State Images Controlled Articulated Object Generation Powered by Diverse Data (NeurIPS 2025 · similarity 0.90)
- CAGE: Controllable Articulation GEneration (CVPR 2024 · similarity 0.89)
- URDF-Anything+: End-to-End Generation for Simulation-Ready Articulated Assets (NeurIPS 2026 · similarity 0.88)
- Real2Code: Reconstruct Articulated Objects via Code Generation (ICLR 2025 · similarity 0.88)
- REACTO: Reconstructing Articulated Objects from a Single Video (CVPR 2024 · 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.
