Research map / 3D vision / Human pose estimation
EquiCaps: Predictor-Free Pose-Aware Pre-Trained Capsule Networks
Athinoulla Konstantinou, Georgios Leontidis, Mamatha Thota, Aiden Durrant
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 |
|---|---|---|---|
| replace predictor-based equivariance with predictor-free capsule pose matrices | EquiCaps · 3DIEBench-T rotation (rotation-only pretraining) | Rotation R2: 0.45 -> 0.73 vs SIE | Improved |
| use EquiCaps instead of invariant-only or predictor-based methods for classification | EquiCaps · 3DIEBench classification | Top-1 accuracy: 83.24 vs 87.44 for SimCLR + AugSelf | Hurt |
| add invariance loss to the equivariance objective (ablation) | EquiCaps · 3DIEBench classification | Top-1 accuracy: 81.43 -> 83.24 | Improved |
| apply variance-covariance regularisation on concatenated activations and poses instead of pose only (ablation) | EquiCaps · 3DIEBench classification | Top-1 accuracy: 81.68 -> 82.40 | Improved |
| increase number of capsules from 16 to 64 (ablation) | EquiCaps · 3DIEBench classification | Top-1 accuracy: 81.89 -> 83.66 | Improved |
| increase capsules from 32 to 64 under rotation-only pretraining (ablation) | EquiCaps · 3DIEBench-T translation | Translation R2: 0.60 -> 0.53 | Hurt |
| add translation to pretraining objective | SIE · 3DIEBench-T translation | Translation R2: 0.20 -> 0.22 | No clear effect |
| use capsule projector rather than MLP predictors on ViT-Tiny/16 backbone | EquiCaps · 3DIEBench rotation (ViT-Tiny/16) | Rotation R2: 0.39 -> 0.71 vs SIE | Improved |
| evaluate on embeddings instead of representations | invariant SSL methods · 3DIEBench rotation | Rotation R2: 0.45-0.52 -> 0.24 | Hurt |
| use EquiCaps for faster convergence | EquiCaps · 3DIEBench rotation | Rotation R2: 0.75 at 100 epochs vs SIE 0.29, CapsIE 0.31 | Improved |
Closest papers by meaning
- Building Deep Equivariant Capsule Networks (ICLR 2020 · similarity 0.91)
- Group Equivariant Capsule Networks (NeurIPS 2018 · similarity 0.90)
- 3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction (NeurIPS 2024 · similarity 0.87)
- Pose-Transformed Equivariant Network for 3D Point Trajectory Prediction (CVPR 2024 · similarity 0.86)
- Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning (ICLR 2025 · similarity 0.86)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · similarity 0.86)
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.
