Research map / Label-efficient and robust learning / Recognition and long-tail
PLAD: Learning To Infer Shape Programs With Pseudo-Labels and Approximate Distributions
Check 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 |
|---|---|---|---|
| Fine-tune with ST instead of REINFORCE | Shape program inference model · 2D CSG test | Chamfer distance (CD): 1.097 -> 0.841 | Helped |
| Fine-tune with LEST instead of REINFORCE | Shape program inference model · 3D CSG test | IoU: 53.4 -> 69.8 | Helped |
| Fine-tune with ST instead of REINFORCE | Shape program inference model · ShapeAssembly test | IoU: 50.8 -> 61.3 | Helped |
| Combine LEST, ST, and WS updates | Shape program inference model · 2D CSG test | Chamfer distance (CD): 0.841 -> 0.811 | Helped |
| Combine LEST, ST, and WS updates | Shape program inference model · 3D CSG test | IoU: 70.8 -> 74.3 | Helped |
| Combine LEST, ST, and WS updates | Shape program inference model · ShapeAssembly test | IoU: 61.3 -> 66.4 | Helped |
| Maintain the best program for each shape across rounds rather than resetting P^BEST (ablation) | ST · 2D CSG test | Chamfer distance (CD): 0.881 -> 0.841 | Helped |
| Maintain the best program for each shape across rounds rather than resetting P^BEST (ablation) | LEST · 2D CSG test | Chamfer distance (CD): 1.011 -> 0.976 | Helped |
| Maintain the best program for each shape across rounds rather than resetting P^BEST (ablation) | LEST+ST · 2D CSG test | Chamfer distance (CD): 0.853 -> 0.829 | Helped |
| Maintain the best program for each shape across rounds rather than resetting P^BEST (ablation) | LEST+ST+WS · 2D CSG test | Chamfer distance (CD): 0.845 -> 0.811 | Helped |
| Fine-tune on CAD shapes with LEST+ST+WS rather than retain supervised pretraining | Shape program inference model · 2D icons (The Noun Project) | average CD: 1.9 -> 4.1 | Hurt |
| Use LEST rather than REINFORCE | Shape program inference model · 2D CSG convergence | time to reach CD 1.097: 36 hours -> 1.1 hours (32x faster) | Helped |
Closest papers by meaning
- Learning to Infer and Execute 3D Shape Programs (ICLR 2019 · similarity 0.89)
- Unsupervised Learning of Shape Programs with Repeatable Implicit Parts (NeurIPS 2022 · similarity 0.87)
- ShapeOdds: Variational Bayesian Learning of Generative Shape Models (CVPR 2017 · similarity 0.86)
- PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction (CVPR 2025 · similarity 0.84)
- ShapeWalk: Compositional Shape Editing Through Language-Guided Chains (CVPR 2024 · similarity 0.84)
- $\texttt{NAISR}$: A 3D Neural Additive Model for Interpretable Shape Representation (ICLR 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.