Research map / 3D vision / 3D generation
DiffuView: Multi-View Diffusion Pretraining for 3D Aware Robotic Manipulation
Kaizhao Zhang, Tian Niu, Tianyu Liu, Chenen Guo, Zijun Xu, Qingda Hu, Wenchao Ding
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 |
|---|---|---|---|
| fine-tune the multi-view diffusion model on robot-centric data (ablation) | DiffuView · LIBERO-10 | success rate (%): 63.3 -> 89.2 | Improved |
| add Plucker ray embedding (ablation) | DiffuView · LIBERO-10 | success rate (%): 76.2 -> 89.2 | Improved |
| add FiLM language conditioning in Q-Former (ablation) | DiffuView · LIBERO-10 | success rate (%): 73.3 -> 89.2 | Improved |
| activate Top-2 experts instead of 1 in noise-conditioned MoE (ablation) | DiffuView · LIBERO-10 | success rate (%): 87.7 -> 89.2 | Improved |
| use multi-view diffusion pretrained encoder instead of OpenVLA baseline under camera rotation | visuomotor policy · Mv-Bench average over viewpoint angles | success rate (%): 39.3 -> 59.2 | Improved |
| large viewpoint shifts beyond the trained range | DiffuView · Mv-Bench 60 degrees | success rate (%): 86.2 (0 deg) -> 34.6 (60 deg) | Hurt |
| multi-view diffusion pretraining vs prior methods on MetaWorld | DiffuView · MetaWorld average (50 tasks) | success rate: 0.682 (VPP) -> 0.706 | Improved |
| DiffuView vs pi0.5-ki on LIBERO-90 | DiffuView · LIBERO-90 | success rate (%): 92.7 -> 92.5 | No clear effect |
| DiffuView vs OpenVLA on real-world tasks | DiffuView · Real-world average (4 tasks) | success rate: 0.63 -> 0.65 | Mixed |
Closest papers by meaning
- PDFactor: Learning Tri-Perspective View Policy Diffusion Field for Multi-Task Robotic Manipulation (CVPR 2025 · similarity 0.91)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.90)
- DynaRend: Learning 3D Dynamics via Masked Future Rendering for Robotic Manipulation (NeurIPS 2025 · similarity 0.90)
- GravMAD: Grounded Spatial Value Maps Guided Action Diffusion for Generalized 3D Manipulation (ICLR 2025 · similarity 0.90)
- Localizing, Structuring, and Rendering: Bridging 3D and 2D Vision-Language-Action Models for Robotic Manipulation (CVPR 2026 · similarity 0.90)
- Generate Subgoal Images before Act: Unlocking the Chain-of-Thought Reasoning in Diffusion Model for Robot Manipulation with Multimodal Prompts (CVPR 2024 · similarity 0.90)
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.
