Research map / Reinforcement learning / Visual and representation RL
Modeling 3D Shapes by Reinforcement Learning
Cheng Lin, Tingxiang Fan, Wenping Wang, Matthias Nießner
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 imitation learning with virtual expert to RL-only training (ablation) | Prim-Agent · ShapeNet Airplane | average accumulated reward: 0.377 -> 0.764 | Improved |
| add imitation learning with virtual expert to RL-only training (ablation) | Mesh-Agent · ShapeNet Airplane | average accumulated reward: -0.013 -> 0.134 | Improved |
| add RL phase to interactive imitation learning (ablation) | Mesh-Agent · ShapeNet Guitar | average accumulated reward: 0.089 -> 0.204 | Improved |
| use interactive guidance instead of fixed demonstrations (ablation) | Mesh-Agent · ShapeNet Airplane | average accumulated reward: 0.019 -> 0.134 | Improved |
| use interactive guidance instead of fixed demonstrations (ablation) | Prim-Agent · ShapeNet Guitar | average accumulated reward: 0.723 -> 0.987 | Improved |
| add double replay buffers to DAgger (ablation) | Prim-Agent · ShapeNet Airplane | average accumulated reward: 0.574 -> 0.725 | Improved |
| add RL on top of DAgger with double replay buffers (ablation) | Prim-Agent · ShapeNet Airplane | average accumulated reward: 0.725 -> 0.764 | Improved |
| full training strategy vs DAgger with double replay buffers (ablation) | Mesh-Agent · ShapeNet Guitar | IoU: 0.381 -> 0.512 | Improved |
| full training strategy vs DAgger with double replay buffers (ablation) | Prim-Agent · ShapeNet Airplane | IoU: 0.131 -> 0.179 | Improved |
| use RGB image reference instead of depth map | Prim-Agent · ShapeNet Airplane | average accumulated reward: 0.764 -> 0.721 | No clear effect |
| use RGB image reference instead of depth map | Mesh-Agent · ShapeNet Airplane | average accumulated reward: 0.134 -> 0.120 | No clear effect |
| use edge loops for mesh editing instead of per-vertex editing (ablation) | Mesh-Agent · ShapeNet | IoU | Improved |
Closest papers by meaning
- DeepMesh: Auto-Regressive Artist-mesh Creation with Reinforcement Learning (ICCV 2025 · similarity 0.88)
- RL-CycleGAN: Reinforcement Learning Aware Simulation-to-Real (CVPR 2020 · similarity 0.86)
- MoDem: Accelerating Visual Model-Based Reinforcement Learning with Demonstrations (ICLR 2023 · similarity 0.86)
- Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards (NeurIPS 2025 · similarity 0.84)
- ReCAD: Reinforcement Learning Enhanced Parametric CAD Model Generation with Vision-Language Models (AAAI 2026 · similarity 0.84)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · 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.
