Research map / 3D vision / Human pose estimation
Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses
Zheng Dang, Kwang Moo Yi, Yinlin Hu, Fei Wang, Pascal Fua, Mathieu Salzmann
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 explicit eigendecomposition-based loss (Essential_Only) with the eigendecomposition-free loss (ablation) | keypoint matching network · SUN3D and Strecha keypoint matching | mAP | Improved |
| add classification-loss pre-training before the essential loss (ablation) | eigendecomposition-free loss · SUN3D and Strecha keypoint matching | mAP | No clear effect |
| use eigendecomposition-free loss instead of SVD/Eigh gradients with vanilla gradient descent (ablation) | plane fitting with one outlier · Plane fitting toy example | iterations to converge | Improved |
| use eigendecomposition-free loss instead of SVD/Eigh gradients with Adam (ablation) | plane fitting with multiple outliers · Plane fitting toy example (multiple outliers) | inlier/outlier recovery | Improved |
| use eigendecomposition-free loss instead of explicit eigendecomposition loss (ablation) | PnP weight network · Synthetic PnP | training convergence | Improved |
| use eigendecomposition-free loss instead of REPPnP as outliers increase | PnP weight network · Synthetic PnP (200 matches, outliers 10 to 150) | rotation error: REPPnP degrades when there are more than 40 outliers; ours near zero error up to 130 outliers | Improved |
| use eigendecomposition-free PnP network instead of EPnP+RANSAC | PnP weight network · PnP real images (Reichstag) | rotation error: 3.1161 -> 0.1319 | Improved |
| use eigendecomposition-free PnP network instead of OPnP | PnP weight network · PnP real images (Florence) | rotation error: 1.5771 -> 0.0416 | Improved |
| use eigendecomposition-free PnP network instead of EPnP+RANSAC | PnP weight network · PnP real images (Notre-dame) | translation error (normalized): 1.0605 -> 0.0046 | Improved |
| use eigendecomposition-free PnP network instead of REPPnP | PnP weight network · PnP real images (Prague) | translation error (normalized): 0.6156 -> 0.0069 | Improved |
Closest papers by meaning
- Eigendecomposition-Free Training of Deep Networks for Linear Least-Square Problems (TPAMI 2020 · similarity 0.97)
- Backpropagation-Friendly Eigendecomposition (NeurIPS 2019 · similarity 0.91)
- Amortized Eigendecomposition for Neural Networks (NeurIPS 2024 · similarity 0.89)
- Batch-Efficient EigenDecomposition for Small and Medium Matrices (ECCV 2022 · similarity 0.87)
- STNet: Spectral Transformation Network for Solving Operator Eigenvalue Problem (NeurIPS 2025 · similarity 0.85)
- Preserving Deep Representations in One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework (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.
