Research map / 3D vision / Human pose estimation
TexPose: Neural Texture Learning for Self-Supervised 6D Object Pose Estimation
Read the paperCheck 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 |
|---|---|---|---|
| Replace RGB-only render-and-compare self-supervision with texture learning and synthetic pose supervision | GDR pose estimator · LineMOD | ADD(-S) average: 77.4% to 91.7% | Helped |
| Use TexPose instead of Self6D++ with depth supervision | Self-supervised pose estimation · Occluded LineMOD | ADD(-S) average: 64.7% vs 66.7% | Helped |
| Adapt using learned realistic textures rather than Self6D++ self-supervision | GDR pose estimator · HomebrewedDB | ADD(-S) average: 86.8% vs. 89.1% | Helped |
| Adapt the texture learner on HomebrewedDB rather than using the LineMOD-trained texture learner | TexPose · HomebrewedDB | ADD(-S) average: 89.1% vs 89.4% | No clear effect |
| Add E_adv to the texture-learning objective (ablation) | TexPose · LineMOD partial objects | ADD(-S) average: 73.3 -> 80.7 | Helped |
| Add E_reg to the texture-learning objective (ablation) | TexPose · LineMOD partial objects | ADD(-S) average: 73.3 -> 83.0 | Helped |
| Use both E_adv and E_reg instead of neither (ablation) | TexPose · LineMOD partial objects | ADD(-S) average: 73.3 -> 84.7 | Helped |
| Add E_adv when E_reg is already used (ablation) | TexPose · LineMOD partial objects | ADD(-S) average: 83.0 -> 84.7 | Helped |
| Add E_reg when E_adv is already used (ablation) | TexPose · LineMOD partial objects | ADD(-S) average: 80.7 -> 84.7 | Helped |
| Increase the self-supervision training set from 9 images (ablation) | TexPose · LineMOD | ADD(-S) score: ca. 18% | Helped |
| Continue texture-learning and pose-learning optimization loops beyond one loop (ablation) | TexPose · LineMOD | ADD(-S) score | No clear effect |
| Optimize poses directly from raw 2D render-and-compare cues instead of supervising with synthesized textured views (ablation) | Self-supervised pose estimation · Duck sequence of LineMOD | pose error | Hurt |
Closest papers by meaning
- Segmentation-Driven 6D Object Pose Estimation (CVPR 2019 · similarity 0.86)
- Feature Mapping for Learning Fast and Accurate 3D Pose Inference From Synthetic Images (CVPR 2018 · similarity 0.86)
- FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024 · similarity 0.86)
- Co-op: Correspondence-based Novel Object Pose Estimation (CVPR 2025 · similarity 0.85)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.85)
- iG-6DoF: Model-free 6DoF Pose Estimation for Unseen Object via Iterative 3D Gaussian Splatting (CVPR 2025 · 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.