Research map / 3D vision / Depth and stereo
EAR-Net: Pursuing End-to-End Absolute Rotations From Multi-View Images
Yuzhen Liu, Qiulei Dong
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 end-to-end training with learned edge weights (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 9.08 -> 4.03 | Improved |
| add end-to-end training (ablation) | EAR-Net · ScanNet | Median rotation error (deg): 3.92 -> 2.06 | Improved |
| add pretraining of feature encoder and rotation branch (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 6.05 -> 4.03 | Improved |
| add learned confidence branch (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 9.31 -> 4.03 | Improved |
| add learned confidence branch (ablation) | EAR-Net · ScanNet | Acc@10deg: 78.71 -> 94.18 | Improved |
| add confidence-aware initialization (CAI) (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 4.83 -> 4.03 | Improved |
| replace l2 loss with Cauchy loss (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 9.31 -> 7.18 | Improved |
| replace l2 loss with Geman-McClure loss (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 9.31 -> 7.36 | Improved |
| use proposed confidence-aware loss instead of Cauchy loss (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 7.18 -> 4.03 | Improved |
| use designed confidence estimation instead of existing confidence regression objective (ablation) | EAR-Net · ScanNet | Mean rotation error (deg): 7.07 -> 4.03 | Improved |
| train Reg6D and RAGO jointly end-to-end instead of two-stage (ablation) | Reg6D+RAGO · ScanNet | Mean rotation error (deg): 11.87 -> 11.57 | Improved |
| train ExtremeRotation and RAGO jointly end-to-end instead of two-stage (ablation) | ExtremeRotation+RAGO · ScanNet | Mean rotation error (deg): 9.83 -> 9.04 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DUST3R — DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- NeuRoRA: Neural Robust Rotation Averaging (ECCV 2020 · similarity 0.84)
- UprightNet: Geometry-Aware Camera Orientation Estimation From Single Images (ICCV 2019 · similarity 0.84)
- RotationNet for Joint Object Categorization and Unsupervised Pose Estimation from Multi-View Images (TPAMI 2019 · similarity 0.84)
- Orient Anything V2: Unifying Orientation and Rotation Understanding (NeurIPS 2025 · similarity 0.83)
- Symmetry-Robust 3D Orientation Estimation (ICML 2025 · similarity 0.82)
- Modeling Stereo-Confidence out of the End-to-End Stereo-Matching Network via Disparity Plane Sweep (AAAI 2024 · similarity 0.82)
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.
