Research map / 3D vision / Face and shape
Neural Shape Parsers for Constructive Solid Geometry
Gopal Sharma, Rishabh Goyal, Difan Liu, Evangelos Kalogerakis, Subhransu Maji
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 execution stack to the encoder (ablation) | CSGNet · Synthetic 2D shapes | IOU (k=1): 86.77 -> 91.33 | Improved |
| add execution stack to the encoder (ablation) | CSGNet · Synthetic 2D shapes | CD (k=10): 0.32 -> 0.12 | Improved |
| increase beam size from 1 to 10 (ablation) | CSGNetStack · Synthetic 2D shapes | IOU: 91.33 -> 93.45 | Improved |
| replace nearest neighbor retrieval with CSGNetStack | NN · Synthetic 2D shapes | IOU (k=1): 73.9 -> 91.33 | Improved |
| train with only 5% of the synthetic data using the stack (ablation) | CSGNet · Synthetic 2D shapes | IOU: 70% -> 80% | Improved |
| fine-tune with RL on CAD training split (ablation) | CSGNetStack k=1 · 2D CAD shapes | CD (i=0): 3.98 -> 1.27 | Improved |
| fine-tune with RL on CAD training split (ablation) | CSGNet k=1 · 2D CAD shapes | CD (i=0): 2.45 -> 1.40 | Improved |
| add visually guided refinement to convergence (ablation) | CSGNetStack RL k=10 · 2D CAD shapes | CD: 1.02 -> 0.34 | Improved |
| add stack to the RL-fine-tuned parser (ablation) | CSGNet RL k=10 · 2D CAD shapes | CD (i=inf): 0.41 -> 0.34 | Improved |
| pretrain with 5% of the synthetic data using the stack (ablation) | CSGNet · 2D CAD shapes | IOU: 46% -> 60% | Improved |
| increase synthetic pretraining beyond 15% (ablation) | CSGNetStack · 2D CAD shapes | IOU | Hurt |
| add stack to the 3D parser (ablation) | 3D-CSGNet · Synthetic 3D shapes | IOU (k=10): 89.2 -> 90.5 | Improved |
Closest papers by meaning
- CSGNet: Neural Shape Parser for Constructive Solid Geometry (CVPR 2018 · similarity 0.96)
- CSG-Stump: A Learning Friendly CSG-Like Representation for Interpretable Shape Parsing (ICCV 2021 · similarity 0.89)
- D$^2$CSG: Unsupervised Learning of Compact CSG Trees with Dual Complements and Dropouts (NeurIPS 2023 · similarity 0.89)
- UCSG-NET- Unsupervised Discovering of Constructive Solid Geometry Tree (NeurIPS 2020 · similarity 0.88)
- Neural Convolutional Surfaces (CVPR 2022 · similarity 0.87)
- Learning Mesh Representations via Binary Space Partitioning Tree Networks (TPAMI 2021 · similarity 0.87)
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.
