Research map / 3D vision / Face and shape
CSGNet: Neural Shape Parser for Constructive Solid Geometry
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 CSGNet instead of nearest-neighbor program retrieval | Synthetic 2D shape parsing · Synthetic 2D test | Chamfer distance (pixels): 1.94 -> 1.39 | Helped |
| Increase beam size from 1 to 5 (ablation) | CSGNet · Synthetic 2D test | Chamfer distance (pixels): 1.69 -> 1.46 | Helped |
| Increase beam size from 5 to 10 (ablation) | CSGNet · Synthetic 2D test | Chamfer distance (pixels): 1.46 -> 1.39 | Helped |
| Use supervised CSGNet with beam size 10 instead of nearest-neighbor retrieval | 2D CAD shape parsing · 2D CAD test | Chamfer distance (pixels): 1.92 -> 1.60 | Helped |
| Add policy-gradient fine-tuning after supervised pretraining (ablation) | CSGNet, beam size 10 · 2D CAD test | Chamfer distance (pixels): 1.60 -> 1.14 | Helped |
| Train with RL alone rather than supervised pretraining followed by RL (ablation) | CSGNet · 2D CAD test | Chamfer distance (pixels) | Hurt |
| Increase beam size from 1 to 10 after RL fine-tuning (ablation) | CSGNet · 2D CAD test | Chamfer distance (pixels): 1.26 -> 1.14 | Helped |
| Add visually guided refinement for 10 iterations (ablation) | CSGNet RL, beam size 10 · 2D CAD test | Chamfer distance (pixels): 1.14 -> 0.42 | Helped |
| Increase beam size from 1 to 10 (ablation) | 3D-CSGNet · Synthetic 3D test | IOU (%): 80.1 -> 89.2 | Helped |
| Use 3D-CSGNet instead of nearest-neighbor retrieval | Synthetic 3D shape parsing · Synthetic 3D test | IOU (%): 73.2 -> 89.2 | Helped |
| Use CSGNet rather than Faster R-CNN for primitive detection | Synthetic 2D primitive detection · Synthetic 2D primitive detection | Mean average precision: 80.1 -> 84.3 | Helped |
| Increase beam size from 10 to 40 for primitive detection (ablation) | CSGNet · Synthetic 2D primitive detection | Mean average precision: 83.0 -> 84.3 | Helped |
Closest papers by meaning
- Learning to Infer and Execute 3D Shape Programs (ICLR 2019 · similarity 0.88)
- Neural Parts: Learning Expressive 3D Shape Abstractions With Invertible Neural Networks (CVPR 2021 · similarity 0.87)
- CoFie: Learning Compact Neural Surface Representations with Coordinate Fields (NeurIPS 2024 · similarity 0.86)
- PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction (CVPR 2025 · similarity 0.85)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.85)
- Geometry-Informed Neural Networks (ICML 2025 · 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.