Research map / 3D vision / Depth and stereo
Taking a Deeper Look at the Inverse Compositional Algorithm
Zhaoyang Lv, Frank Dellaert, James M. Rehg, Andreas Geiger
Check 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 the two-view feature encoder to inverse compositional alignment (ablation) | Deep IC on MovingObjects3D · MovingObjects3D Small | 3D EPE (cm): No Learning 11.26 -> Ours: (A) 3.66 | Improved |
| Add the convolutional M-estimator (ablation) | Deep IC with feature encoder A on BundleFusion · BundleFusion Small | 3D EPE (cm): 2.65 -> 1.75 | Improved |
| Add the trust-region network (ablation) | Deep IC with modules A and B on DynamicBundleFusion · DynamicBundleFusion Large | 3D EPE (cm): 6.88 -> 4.64 | Improved |
| Remove shared weights across pyramid scales (ablation) | Full Deep IC on DynamicBundleFusion · DynamicBundleFusion Large | 3D EPE (cm): 4.64 -> 3.82 | Improved |
| Reduce inner iterations per scale to K=1 (ablation) | Full Deep IC on MovingObjects3D · MovingObjects3D Small | 3D EPE (cm): 2.91 -> 3.37 | Hurt |
| Increase inner iterations per scale to K=5 (ablation) | Full Deep IC on MovingObjects3D · MovingObjects3D Small | 3D EPE (cm): 2.91 -> 2.92 | No clear effect |
| Add the trust-region network (ablation) | Deep IC with modules A and B on the 2D affine task · 2D affine test set | L1 error: 0.132 -> 0.071 | Improved |
| Use the full learned method instead of RGB-D visual odometry | TUM RGB-D SLAM · TUM RGB-D SLAM test split, KF 8 | mRPE translation (cm): 13.83 -> 5.88 | Improved |
Closest papers by meaning
- Compositional Human Pose Regression (ICCV 2017 · similarity 0.86)
- The Conditional Lucas & Kanade Algorithm (ECCV 2016 · similarity 0.86)
- Fast Optical Flow Using Dense Inverse Search (ECCV 2016 · similarity 0.85)
- SuperPrimitive: Scene Reconstruction at a Primitive Level (CVPR 2024 · similarity 0.84)
- MP-SfM: Monocular Surface Priors for Robust Structure-from-Motion (CVPR 2025 · similarity 0.84)
- PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting for Novel View Synthesis (ICML 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.
