Research map / 3D vision / Human pose estimation
VI-Net: Boosting Category-level 6D Object Pose Estimation via Learning Decoupled Rotations on the Spherical Representations
Jiehong Lin, Zewei Wei, Yabin Zhang, Kui Jia
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 |
|---|---|---|---|
| decompose rotation into viewpoint and in-plane rotations (ablation) | VI-Net · REAL275 | mAP 5°2cm: 45.0 -> 50.0 | Improved |
| use binary classification on the sphere instead of direct regression for viewpoint rotation (ablation) | VI-Net V-Branch · REAL275 | mAP 5°2cm: 43.1 -> 49.3 | Improved |
| split classification into separate azimuth and inclination classifiers (ablation) | VI-Net V-Branch · REAL275 | mAP 5°2cm: 49.3 -> 50.0 | Improved |
| add feature transformation to the canonical zenith view in I-Branch (ablation) | VI-Net · REAL275 | mAP 5°2cm: 33.7 -> 50.0 | Improved |
| use SPA-SConv instead of SPE-SConv spectral convolution (ablation) | VI-Net · REAL275 | mAP 5°2cm: 35.1 -> 50.0 | Improved |
| add feature padding to SPA-SConv (ablation) | VI-Net · REAL275 | mAP 5°2cm: 48.5 -> 50.0 | Improved |
| add symmetric (flip and max) convolution operation (ablation) | SPA-SConv · REAL275 | mAP 5°2cm: 46.9 -> 48.5 | Improved |
| include translation and size estimation inside VI-Net instead of separate PN2 (ablation) | VI-Net · REAL275 | mAP 5°2cm: 50.0 -> 45.0 | Hurt |
| add RGB input in addition to depth (ablation) | VI-Net · REAL275 | mAP 5°2cm: 43.0 -> 50.0 | Improved |
| compare VI-Net (no shape priors) with shape-prior methods on CAMERA25 | VI-Net · CAMERA25 | mAP 5°2cm: VI-Net 74.1 vs SPD+CATRE 75.4 | Mixed |
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.
- DualPoseNet — DualPoseNet: Category-Level 6D Object Pose and Size Estimation Using Dual Pose Network With Refined Learning of Pose Consistency (ICCV 2021) · better on 1 of 1 shared result
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
- FS-Net: Fast Shape-Based Network for Category-Level 6D Object Pose Estimation With Decoupled Rotation Mechanism (CVPR 2021 · similarity 0.91)
- G2L-Net: Global to Local Network for Real-Time 6D Pose Estimation With Embedding Vector Features (CVPR 2020 · similarity 0.88)
- RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews From Unsupervised Viewpoints (CVPR 2018 · similarity 0.88)
- RotationNet for Joint Object Categorization and Unsupervised Pose Estimation from Multi-View Images (TPAMI 2019 · similarity 0.88)
- One2Any: One-Reference 6D Pose Estimation for Any Object (CVPR 2025 · similarity 0.85)
- 3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction (NeurIPS 2024 · similarity 0.85)
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.
