Research map / 3D vision / Face and shape
Representation Learning on Unit Ball with 3D Roto-translational Equivariance
Sameera Ramasinghe, Salman Hameed Khan, Nick M. Barnes, Stephen Jay Gould
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 more convolution layers (2 or 3) (ablation) | volumetric conv net · ModelNet10 | accuracy: 93.8% -> 92.0% and 89.8% | Hurt |
| use volumetric convolution instead of spherical convolution (ablation) | FA · ModelNet10 | accuracy: 76.2% -> 93.8% | Improved |
| add adaptive frequency pooling instead of mean pooling (ablation) | FA · ModelNet10 | accuracy: 84.2% -> 93.8% | Improved |
| add adaptive frequency pooling instead of max pooling (ablation) | FA · ModelNet10 | accuracy: 86.7% -> 93.8% | Improved |
| add adaptive frequency pooling instead of feature concatenation (ablation) | FA · ModelNet10 | accuracy: 87.8% -> 93.8% | Improved |
| add adaptive frequency pooling instead of MIL attention pooling (ablation) | FA · ModelNet10 | accuracy: 90.3% -> 93.8% | Improved |
| add adaptive frequency pooling instead of MIL gated attention pooling (ablation) | FA · ModelNet10 | accuracy: 89.8% -> 93.8% | Improved |
| use angular weighted-softmax loss instead of softmax (ablation) | FA · ModelNet10 | accuracy: 93.8% -> 92.8% | Hurt |
| use generalized angular softmax loss instead of softmax (ablation) | FA · ModelNet10 | accuracy: 93.8% -> 90.7% | Hurt |
| use cosine similarity instead of Euclidean distance for retrieval (ablation) | FA · SHREC'17 | mAP: 0.386 -> 0.452 | Improved |
| use cosine similarity instead of KL divergence for retrieval (ablation) | FA · SHREC'17 | mAP: 0.320 -> 0.452 | Improved |
| use cosine similarity instead of Bhattacharyya distance for retrieval (ablation) | FA · SHREC'17 | mAP: 0.354 -> 0.452 | Improved |
Closest papers by meaning
- Deep Learning 3D Shapes Using Alt-az Anisotropic 2-Sphere Convolution (ICLR 2019 · similarity 0.86)
- Volumetric and Multi-View CNNs for Object Classification on 3D Data (CVPR 2016 · similarity 0.85)
- Spin-Weighted Spherical CNNs (NeurIPS 2020 · similarity 0.85)
- Scalable and Equivariant Spherical CNNs by Discrete-Continuous (DISCO) Convolutions (ICLR 2023 · similarity 0.85)
- Cross-Domain 3D Equivariant Image Embeddings (ICML 2019 · similarity 0.84)
- On the Fourier analysis in the SO(3) space : the EquiLoPO Network (ICLR 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.
