Research map / 3D vision / Human pose estimation
Neural Object Learning for 6D Pose Estimation Using a Few Cluttered Images
Kiru Park, Timothy Patten, Markus Vincze
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 pose refinement of source images (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.181 -> 0.173 | Improved |
| add decoder block (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.289 -> 0.173 | Improved |
| add LSTM module instead of simple average (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.241 -> 0.173 | Improved |
| add perceptual loss to L1 loss (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.184 -> 0.177 | Improved |
| add smooth loss to L1 and perceptual loss (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.177 -> 0.173 | Improved |
| add adversarial GAN loss (ablation) | NOL · SMOT rendered image quality | perceptual similarity (lower is better): 0.173 -> 0.183 | Hurt |
| train Pix2Pose on NOL images instead of synthetic images from 3D models | Pix2Pose · LineMOD RGB | ADD average: 16.6 -> 49.5 | Improved |
| train on NOL images instead of synthetic images, with ICP | Pix2Pose + ICP · LineMOD RGB+ICP | ADD average: 78.8 -> 96.3 | Improved |
| train on NOL images instead of synthetic images | Pix2Pose · LineMOD-Occ RGB | BOP score: 20.0 -> 37.7 | Improved |
| add ICP3D refinement with NOL training | Pix2Pose · LineMOD-Occ | BOP score: 37.7 -> 61.3 | Improved |
| train on NOL images instead of real images (precise models) | Pix2Pose · SMOT RGB | AD{D|I} score: 25.0 -> 35.5 | Improved |
| render NOL images with NOL instead of G2Ltex textures | Pix2Pose · SMOT RGB | AD{D|I} score: 25.7 -> 35.5 | Improved |
Closest papers by meaning
- TexPose: Neural Texture Learning for Self-Supervised 6D Object Pose Estimation (CVPR 2023 · similarity 0.91)
- FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024 · similarity 0.88)
- Any6D: Model-free 6D Pose Estimation of Novel Objects (CVPR 2025 · similarity 0.88)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.88)
- One2Any: One-Reference 6D Pose Estimation for Any Object (CVPR 2025 · similarity 0.88)
- Open-Vocabulary Object 6D Pose Estimation (CVPR 2024 · similarity 0.88)
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.
