Research map / Point clouds / Shape and surface completion
BAE-NET: Branched Autoencoder for Shape Co-Segmentation
Zhiqin Chen, Kangxue Yin, Matthew Fisher, Siddhartha Chaudhuri, Hao Zhang
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 table shapes to chair training set (joint chair+table) (ablation) | BAE-NET unsupervised · ShapeNet part chair | IOU: 65.5 -> 83.7 | Improved |
| allow part combinations when scoring (mod-IOU instead of IOU) | BAE-NET unsupervised · ShapeNet part airplane | IOU: 61.1 -> 80.4 | Improved |
| allow part combinations when scoring (mod-IOU instead of IOU) | BAE-NET unsupervised · ShapeNet part skateboard | IOU: 63.5 -> 88.1 | Improved |
| allow part combinations when scoring (mod-IOU instead of IOU) | BAE-NET unsupervised · ShapeNet part table | IOU: 78.7 -> 87.0 | Improved |
| use 3-layer decoder instead of 2-layer decoder (ablation) | BAE-NET · synthetic 2D elements and triple rings | qualitative part separation | Improved |
| use 3-layer decoder instead of 4-layer decoder (ablation) | BAE-NET · synthetic 2D shapes | qualitative part separation | Improved |
| use branched fully connected decoder instead of CNN decoder (ablation) | BAE-NET · synthetic 2D shapes | qualitative part separation | Improved |
| alter data distribution so only 20% of shapes have target part (ablation) | BAE-NET · Tags2Parts datasets | qualitative part separation | Improved |
| use BAE-NET one-shot instead of Tags2Parts | Tags2Parts weak supervision · Tags2Parts engine | AUC: 0.46 -> 0.88 | Improved |
| use BAE-NET with 1 exemplar instead of PointNet++ on 10% data | PointNet++ · ShapeNet part 15 categories | average IOU: 73.5 -> 76.6 | Improved |
| use BAE-NET with 3 exemplars instead of PointNet++ on 30% data | PointNet++ · ShapeNet part 15 categories | average IOU: 76.6 -> 78.7 | Improved |
| reduce capacity and add regularization to supervised baselines (ablation) | PointNet, PointNet++, PointCNN, SSCN · ShapeNet part 1-exemplar vs 10% | average IOU: PointNet 72.1 -> 73.8; PointNet++ 73.5 -> 74.0; PointCNN 58.0 -> 58.6; SSCN 56.7 -> 57.8 | No clear effect |
Later papers that compared with this method
3 later accepted papers report BAE-Net as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Learning Implicit Functions for Dense 3D Shape Correspondence of Generic Objects (TPAMI 2023 · better on 8 of 10 shared results)
- Unsupervised Learning of Intrinsic Structural Representation Points (CVPR 2020 · better on 5 of 15 shared results)
- Learning Implicit Functions for Topology-Varying Dense 3D Shape Correspondence (NeurIPS 2020 · better on 10 of 10 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PointCNN — PointCNN: Convolution On X-Transformed Points (NeurIPS 2018) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AdaCoSeg: Adaptive Shape Co-Segmentation With Group Consistency Loss (CVPR 2020 · similarity 0.88)
- PartNet: A Recursive Part Decomposition Network for Fine-Grained and Hierarchical Shape Segmentation (CVPR 2019 · similarity 0.88)
- RIM-Net: Recursive Implicit Fields for Unsupervised Learning of Hierarchical Shape Structures (CVPR 2022 · similarity 0.88)
- Symmetric Shape-Preserving Autoencoder for Unsupervised Real Scene Point Cloud Completion (CVPR 2023 · similarity 0.88)
- ShapeEmbed: a self-supervised learning framework for 2D contour quantification (NeurIPS 2025 · similarity 0.87)
- ShapeMatcher: Self-Supervised Joint Shape Canonicalization Segmentation Retrieval and Deformation (CVPR 2024 · 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.
