Research map / Video / Video generation
ShapeGen4D: Towards High Quality 4D Shape Generation from Videos
Jiraphon Yenphraphai, Ashkan Mirzaei, Jianqi Chen, Jiaxu Zou, Sergey Tulyakov, Raymond A. Yeh, Peter Wonka, Chaoyang Wang
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 aligned latents (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | Chamfer distance: 0.1348 -> 0.1096 | Improved |
| add shared noise across frames (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | F-Score: 0.2962 -> 0.3190 | Improved |
| add shared noise across frames (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | Chamfer distance: 0.1186 -> 0.1096 | Hurt |
| use full spatiotemporal attention instead of 1D temporal attention (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | Chamfer distance: 0.2118 -> 0.1096 | Improved |
| add image hidden states in spatiotemporal attention (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | F-Score: 0.3084 -> 0.3190 | Improved |
| add denoise time shift (ablation) | ShapeGen4D · Objaverse held-out ablation (8 frames) | Chamfer distance: 0.1374 -> 0.1096 | Improved |
| add shared noise to per-frame Step1X-3D | Step1X-3D · Objaverse held-out test (33 samples) | IoU: 0.3033 -> 0.3149 | Improved |
| add shared noise to per-frame Step1X-3D | Step1X-3D · Objaverse held-out test (33 samples) | Chamfer distance: 0.1356 -> 0.1368 | No clear effect |
| use ShapeGen4D instead of V2M4 (optimization-based) | V2M4 · Objaverse held-out test (33 samples) | Chamfer distance: 0.1233 -> 0.1220 | Improved |
| use ShapeGen4D instead of GVFD | GVFD · Objaverse held-out test (33 samples) | Chamfer distance: 0.3978 -> 0.1220 | Improved |
| use L4GM instead of ShapeGen4D for rendering | ShapeGen4D · Consistent4D (20 videos) | LPIPS: 0.1359 -> 0.0988 | Hurt |
| use ShapeGen4D instead of Step1X-3D per frame | Step1X-3D · Consistent4D (20 videos) | FVD: 940 -> 796 | Improved |
Closest papers by meaning
- SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025 · similarity 0.90)
- Diffusion4D: Fast Spatial-temporal Consistent 4D generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.90)
- 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Restage4D: Reanimating Deformable 3D Reconstruction from a Single Video (NeurIPS 2025 · similarity 0.89)
- Towards Physical Understanding in Video Generation: A 3D Point Regularization Approach (NeurIPS 2025 · similarity 0.89)
- Vidu4D: Single Generated Video to High-Fidelity 4D Reconstruction with Dynamic Gaussian Surfels (NeurIPS 2024 · similarity 0.89)
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.
