Research map / 3D vision / Face and shape
PDO-s3DCNNs: Partial Differential Operator Based Steerable 3D CNNs
Read the paperCheck 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 |
|---|---|---|---|
| Use Gaussian discretization with a 5×5×5 kernel instead of FD with a 3×3×3 kernel (ablation) | SO(3)-steerable PDO-s3DCNN · 3D Tetris, arbitrary rotations | test accuracy (%): 18.20\pm 3.13 -> 99.04\pm 0.14 | Helped |
| Increase the Gaussian kernel size from 3×3×3 to 5×5×5 (ablation) | SO(3)-steerable PDO-s3DCNN · 3D Tetris, arbitrary rotations | test accuracy (%): 29.99\pm 4.98 -> 99.04\pm 0.14 | Helped |
| Use more extensive rotation equivariance, O rather than V or T (ablation) | PDO-s3DCNN with regular features · SHREC'17 perturbed | score: 52.7 -> 58.6; 57.6 -> 58.6 | Helped |
| Use I rather than O as the equivariance group (ablation) | PDO-s3DCNN with regular features · SHREC'17 perturbed | score: 58.6 -> 55.5 | Hurt |
| Use SO(3) irreducible features rather than O regular features (ablation) | PDO-s3DCNN · SHREC'17 perturbed | score: 58.6 vs. 57.4; 58.3 vs. 58.6 | Mixed |
| Use irrep-compatible special BN and nonlinearities instead of standard ones (ablation) | O-steerable PDO-s3DCNN · SHREC'17 perturbed | score: 58.6 -> 56.5 | Hurt |
| Use V-quotient features instead of regular features (ablation) | O-steerable PDO-s3DCNN · SHREC'17 perturbed | score: 58.6 -> 55.5 | Hurt |
| Add V-quotient features alongside regular features (ablation) | O-steerable PDO-s3DCNN · SHREC'17 perturbed | score: 55.5 -> 59.1 | Helped |
| Use O rather than V as the equivariance group (ablation) | PDO-s3DCNN · ISBI 2012 Challenge | V_rand: 0.98415 -> 0.98727 | Helped |
| Replace CubeNet filters with PDO-based equivariant filters (ablation) | V-steerable segmentation model · ISBI 2012 Challenge | V_rand: 0.98018 -> 0.98415 | Helped |
| Use O-steerable PDO-s3DCNN instead of SFCNN | ISBI 2012 Challenge segmentation · ISBI 2012 Challenge | V_rand: 0.98680 -> 0.98727 | Helped |
Closest papers by meaning
- Learning Steerable Filters for Rotation Equivariant CNNs (CVPR 2018 · similarity 0.89)
- FILTRA: Rethinking Steerable CNN by Filter Transform (ICML 2021 · similarity 0.88)
- Clifford-Steerable Convolutional Neural Networks (ICML 2024 · similarity 0.86)
- Learning with 3D rotations, a hitchhiker's guide to SO(3) (ICML 2024 · similarity 0.85)
- Steerable Transformers for Volumetric Data (ICML 2025 · similarity 0.85)
- On the Fourier analysis in the SO(3) space : the EquiLoPO Network (ICLR 2025 · 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.