Research map / 3D vision / 3D generation
Neural Re-Simulation for Generating Bounces in Single Images
Carlo Innamorati, Bryan Russell, Danny M. Kaufman, Niloy J. Mitra
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 the trajectory update network (ablation) | Ours without it (DepNet) · SUNCG synthetic | L2 2D (x1e4): 101.2 -> 1.5 | Improved |
| add the depth correction network (ablation) | Ours without it (TrajNet) · SUNCG synthetic | L2 2D (x1e4): 3.3 -> 1.5 | Improved |
| add the depth correction network (ablation) | Ours without it (TrajNet) · SUNCG synthetic | perceptual loss (x1e2): 2.4 -> 1.3 | Improved |
| add the depth correction network (ablation) | Ours without it (TrajNet) · SUNCG synthetic | L2 3D (x1e4): 4.3 -> 2.1 | Improved |
| add the trajectory update network (ablation) | Ours without it (DepNet) · SUNCG synthetic | perceptual loss (x1e2): 2.9 -> 1.3 | Improved |
| add depth correction on top of depth + forward simulation | Depth + fwdS. · SUNCG synthetic | time to first bounce difference: 3.0 -> 1.8 | Improved |
| histogram-equalize depth before forward simulation | Depth + fwdS. · SUNCG synthetic | L2 2D (x1e4): 186.5 -> 144.1 | Improved |
| use learned neural re-simulation | Depth + fwdS. · real images user study | user preference (%): 71 | Improved |
| add trajectory update network on top of depth correction (ablation) | DepNet · real images user study | user preference (%): 59 | Improved |
| add depth correction network (ablation) | Depth + fwdS. · real images user study | user preference (%): 63 | Improved |
| replace active-sensing depth simulation with neural re-simulation | forward simulation on NYUv2 depth · NYUv2 user study | user preference (%): 49 | No clear effect |
| add multi-bounce and roll-out trajectories (ablation) | Ours with free-fall extension · NYUv2 user study | user preference (%): 56 -> 59 | Improved |
Closest papers by meaning
- Learning rigid-body simulators over implicit shapes for large-scale scenes and vision (NeurIPS 2024 · similarity 0.85)
- PISA Experiments: Exploring Physics Post-Training for Video Diffusion Models by Watching Stuff Drop (ICML 2025 · similarity 0.85)
- InterDyn: Controllable Interactive Dynamics with Video Diffusion Models (CVPR 2025 · similarity 0.85)
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.85)
- Coarse-to-Real: Generative Rendering for Populated Dynamic Scenes (NeurIPS 2026 · similarity 0.85)
- NIFTY: Neural Object Interaction Fields for Guided Human Motion Synthesis (CVPR 2024 · 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.
