Research map / 3D vision / 3D generation
Tangent Space Backpropagation for 3D Transformation Groups
Check 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 tangent-space backpropagation instead of untuned embedding-space Autograd | Inverse kinematics with SO(3) · Inverse kinematics, SO(3) | convergence rate: 0.0 -> 100.0 | Helped |
| Tune the embedding-space implementation for numerical stability | PyTorch+Autograd · Inverse kinematics, SO(3) | convergence rate: 0.0 -> 99.8 | Helped |
| Use tangent-space gradients for pose graph initialization | Pose graph optimization · Parking-Garage | time (seconds): 16.5 -> 1.18 | Helped |
| Use tangent-space backpropagation rather than embedding-space Autograd | Pose graph optimization · Parking-Garage | error: $6.35\times10^{-1}$ -> $6.35\times10^{-1}$ | No clear effect |
| Use a direct geodesic pose loss instead of the indirect optical-flow proxy loss | DeepV2D (ours) · RGB-D SLAM benchmark avg | ATE rmse [m]: 0.113 -> 0.105 | Helped |
| Increase the Sim(3) Jacobian series approximation order (ablation) | RGB-D registration · RGB-D registration, Sim(3) scale | scale (<1%): $98.1\pm0.43$; $98.3\pm0.35$; $97.2\pm0.25$ | No clear effect |
Closest papers by meaning
- Auto-Encoding Transformations in Reparameterized Lie Groups for Unsupervised Learning (AAAI 2021 · similarity 0.86)
- Lie Group Decompositions for Equivariant Neural Networks (ICLR 2024 · similarity 0.83)
- Building Transformation Layers for Riemannian Neural Networks (NeurIPS 2026 · similarity 0.83)
- PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction (CVPR 2025 · similarity 0.83)
- Matrix Manifold Neural Networks++ (ICLR 2024 · similarity 0.83)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · 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.