Research map / 3D vision / 3D generation
Bounce and Learn: Modeling Scene Dynamics with Real-World Bounces
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the PointNet-based point-cloud trajectory encoder instead of center encoding | Bounce and Learn · Bounce Dataset test set | Dist (cm): 23.1 -> 21.3 | Helped |
| Add sensor-estimated normal supervision during training (ablation) | Bounce and Learn · Bounce Dataset test set | % Normals within 30° of estimated normal: 24.08 -> 50.14 | Helped |
| Add sensor-estimated normal supervision during training (ablation) | Bounce and Learn · Bounce Dataset test set | Dist (cm): 21.3 -> 22.7 | Mixed |
| Provide both sensor-estimated normals and COR at test time instead of either alone | Bounce and Learn · Bounce Dataset test set | Post-bounce distance (cm): 21.4 -> 20.6 | Helped |
| Fine-tune PIM on real data instead of fixing it after synthetic-only pretraining (ablation) | Bounce and Learn · Bounce Dataset validation set | Dist (cm): 37.2 -> 19.9 | Helped |
| Fine-tune both the core physics engine and trajectory encoder rather than fixing the encoder (ablation) | Bounce and Learn · Bounce Dataset validation set | % Normals within 30° of estimated normal: 34.98 -> 13.01 | Hurt |
| Add spatial regularization (ablation) | Bounce and Learn · Bounce Dataset validation set | Dist (cm): 20.5 -> 19.9 | Helped |
| Add spatial regularization (ablation) | Bounce and Learn · Bounce Dataset validation set | COR median absolute error: 0.127 -> 0.123 | Helped |
| Use projected ball-center impact locations instead of human annotations (ablation) | Bounce and Learn · Bounce Dataset test set | Dist (cm): 21.3 -> 21.8 | No clear effect |
Closest papers by meaning
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.86)
- FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity (CVPR 2025 · similarity 0.86)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.85)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.85)
- DreamPhysics: Learning Physics-Based 3D Dynamics with Video Diffusion Priors (AAAI 2025 · similarity 0.85)
- Learning 3D Particle-based Simulators from RGB-D Videos (ICLR 2024 · similarity 0.85)
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.