Research map / Graph learning / Graph self-supervised learning
PP-Brep: Few-Shot B-rep Classification with Hybrid Graph Representation
Jiacheng Hao, Chunying Liu, Hao Guo, Ruohan Wang, Hongping Gan, Yilei Shi
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 local topology graph to node-only features (ablation) | PP-Brep · FabWave-31 (1-shot) | accuracy (%): 51.43 -> 57.08 | Improved |
| add global parallel graph (ablation) | PP-Brep · FabWave-31 (1-shot) | accuracy (%): 57.08 -> 60.95 | Improved |
| add regional association hypergraph (ablation) | PP-Brep · FabWave-31 (1-shot) | accuracy (%): 60.95 -> 72.89 | Improved |
| add RL-controlled adaptive perturbation to SimGRACE (ablation) | contrastive pre-training · FabWave-31 (1-shot) | accuracy (%): 67.50 -> 72.89 | Improved |
| use contrastive instead of generative pre-training (GraphMAE2) (ablation) | pre-training · FabWave-31 (1-shot) | accuracy (%): 45.76 -> 59.66 (GraphCL) | Improved |
| use model-perturbation SimGRACE instead of graph-augmentation GraphCL (ablation) | pre-training · FabWave-31 (1-shot) | accuracy (%): 59.66 -> 67.50 | Improved |
| use structure-aware prompts instead of full fine-tuning (ablation) | downstream tuning · FabWave-31 (1-shot) | accuracy (%): 43.82 -> 72.89 | Improved |
| add structural context to prompt (SAP vs non-context prompt) (ablation) | graph prompt · FabWave-31 (1-shot) | accuracy (%): 70.30 -> 72.89 | Improved |
| add structural context to prompt on AUROC (ablation) | graph prompt · FabWave-31 (1-shot) | AUROC (%): 92.00 -> 91.53 | No clear effect |
| use in-place perturbation instead of full backbone duplication (ablation) | contrastive pre-training · FabWave-31 (1-shot) | accuracy (%): 72.39 -> 72.89 | No clear effect |
| use in-place perturbation instead of duplication (ablation) | contrastive pre-training · DeepCAD, batch size 8192 | time per epoch (s): 1104.23 -> 19.31 | Improved |
| increase shots from 1 to 5 | PP-Brep · FabWave-31 | accuracy (%): 72.98 -> 87.55 | Improved |
Closest papers by meaning
- Self-Supervised Representation Learning for CAD (CVPR 2023 · similarity 0.88)
- BrepGiff: Lightweight Generation of Complex B-rep with 3D GAT Diffusion (CVPR 2025 · similarity 0.87)
- BrepVGAE: Variational Graph Autoencoder with Unified Latent Representation for B-rep (CVPR 2026 · similarity 0.86)
- SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation (NeurIPS 2025 · similarity 0.84)
- DTGBrepGen: A Novel B-rep Generative Model through Decoupling Topology and Geometry (CVPR 2025 · similarity 0.84)
- Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph (AAAI 2025 · similarity 0.83)
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.
