Research map / 3D vision / 3D generation
Stable-Sim2Real: Exploring Simulation of Real-Captured 3D Data with Two-Stage Depth Diffusion
Mutian Xu, Chongjie Ye, Haolin Liu, Yushuang Wu, Jiahao Chang, Xiaoguang Han
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 residual depth generation (instead of direct real depth) (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 26.4/18.2/22.1 -> 28.5/16.1/24.7 | Improved |
| add loss re-weighting in Stage-II (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 27.1/17.9/23.3 -> 28.5/16.1/24.7 | Improved |
| add 3D binary discriminator (instead of difference threshold) (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 27.3/17.5/23.6 -> 28.5/16.1/24.7 | Improved |
| add Stage-III on top of Stage-II (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 28.5/16.1/24.7 -> 28.6/16.1/24.5 | No clear effect |
| add Stage-IV on top of Stage-II (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 28.5/16.1/24.7 -> 28.0/16.9/24.1 | Hurt |
| use patch size 128x128 instead of 32x32 (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 27.1/16.9/22.6 -> 28.5/16.1/24.7 | Improved |
| use classifier width 64 instead of 16 (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 26.3/17.2/22.8 -> 28.5/16.1/24.7 | Improved |
| use classifier width 64 instead of 256 (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 26.4/18.3/22.5 -> 28.5/16.1/24.7 | Improved |
| use fixed weights instead of classifier confidence as soft weight (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 27.8/15.2/23.9 -> 28.5/16.1/24.7 | Improved |
| finetune 2D Stable Diffusion instead of 3D diffusion DiT-3D (ablation) | Stable-Sim2Real · LASA 3D reconstruction, table category, zero-shot ablation | mIoU/Chamfer/F-score: 26.9/19.1/22.3 -> 28.5/16.1/24.7 | Improved |
| add Stage-II enhancement to Stage-I output (ablation) | Stable-Sim2Real · 3D binary classifier on simulated patches | classifier accuracy: 67.8% -> 32.5% | Improved |
| use Cycle-GAN simulated data for pretraining instead of none (ablation) | DisCo · LASA 3D reconstruction, few-shot (Chair) | mIoU: 24.5 -> 25.1 | No clear effect |
Closest papers by meaning
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.87)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.87)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.86)
- Text-to-3D Generation with Bidirectional Diffusion using both 2D and 3D priors (CVPR 2024 · similarity 0.86)
- CAT3D: Create Anything in 3D with Multi-View Diffusion Models (NeurIPS 2024 · similarity 0.86)
- BLURD: Benchmarking and Learning using a Unified Rendering and Diffusion Model (NeurIPS 2024 · similarity 0.86)
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.
