Research map / 3D vision / Face and shape
Directionally Convolutional Networks for 3D Shape Segmentation
Haotian Xu, Ming Dong, Zichun Zhong
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 |
|---|---|---|---|
| use directional (curvature-ordered) convolution instead of non-directional convolution with shuffled faces (ablation) | DCN · Human shapes (12 train, 8 test) | train and test logloss | Improved |
| fuse the two streams late by element-wise product instead of at penultimate layers (ablation) | two-stream DCN · PSB + COSEG small categories (average) | segmentation accuracy: 0.9496 -> 0.9522 | Improved |
| late fusion instead of early fusion (ablation) | two-stream DCN · Chairs (400 meshes) | segmentation accuracy: 0.9518 -> 0.9573 | Improved |
| late fusion instead of early fusion (ablation) | two-stream DCN · Vases (300 meshes) | segmentation accuracy: 0.9034 -> 0.9086 | Improved |
| late fusion instead of early fusion (ablation) | two-stream DCN · Mech category | segmentation accuracy: 0.9698 -> 0.9660 | Hurt |
| late fusion instead of early fusion (ablation) | two-stream DCN · Candelabra category | segmentation accuracy: 0.9603 -> 0.9546 | Hurt |
| add global distance-histogram stream to the local DCN stream (ablation) | DCN · t-SNE of Teddy and Ant feature maps | cluster separation | Improved |
| DCN two-stream versus Guo et al. on Bust | mesh segmentation · Bust | segmentation accuracy: 0.6984 -> 0.7898 | Improved |
| DCN two-stream versus ShapePFCN on Mech | mesh segmentation · Mech | segmentation accuracy: 0.9790 -> 0.9660 | Hurt |
| DCN two-stream versus Guo et al. on Chairs | mesh segmentation · Chairs (400 meshes) | segmentation accuracy: 0.9252 -> 0.9573 | Improved |
Closest papers by meaning
- 3D Shape Segmentation With Projective Convolutional Networks (CVPR 2017 · similarity 0.90)
- PartNet: A Recursive Part Decomposition Network for Fine-Grained and Hierarchical Shape Segmentation (CVPR 2019 · similarity 0.89)
- Embedding 3D Geometric Features for Rigid Object Part Segmentation (ICCV 2017 · similarity 0.88)
- SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (CVPR 2017 · similarity 0.87)
- FeaStNet: Feature-Steered Graph Convolutions for 3D Shape Analysis (CVPR 2018 · similarity 0.87)
- Learning 3D Keypoint Descriptors for Non-Rigid Shape Matching (ECCV 2018 · 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.
