Research map / 3D vision / Human pose estimation
Pose Guided RGBD Feature Learning for 3D Object Pose Estimation
Vassileios Balntas, Andreas Doumanoglou, Caner Sahin, Juil Sock, Rigas Kouskouridas, Tae-Kyun Kim
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-guided term Lpose with Lobject in place of Ltriplets + Lpairs | feature learning on LINEMOD · LINEMOD (synthetic training, average) | pose accuracy (%): 30.34 -> 54.32 | Improved |
| add pose-guided term without bootstrapping versus Wohlhart without bootstrapping | feature learning · LINEMOD (synthetic training, average) | pose accuracy (%): 18.61 -> 51.02 | Improved |
| add symmetry weight phi to pose loss (ablation) | Lobject + Lpose · LINEMOD (synthetic training, average) | pose accuracy (%): 51.02 -> 54.32 | Improved |
| add regression loss Lreg (ablation) | Lobject + Lpose phi · LINEMOD (synthetic training, average) | pose accuracy (%): 54.32 -> 55.64 | Improved |
| add regression loss Lreg for unseen objects (protocol 3) (ablation) | Lobject + Lpose · LINEMOD real data, unseen objects | mean angle error | Hurt |
| add regression loss with dissimilar train/test poses (protocol 2) (ablation) | Lobject + Lpose · LINEMOD real data, dissimilar poses | mean angle error | No clear effect |
| add regression loss with random split (protocol 1) (ablation) | Lobject + Lpose · LINEMOD real data, random split | mean angle error | Improved |
| add occlusion term Locc (belt object) (ablation) | pose-guided features · Hand-Object occlusion, belt object | average angle error: 14.3 -> 11.8 | Improved |
| add occlusion term Locc (10 objects) (ablation) | pose-guided features · Hand-Object occlusion, 10 objects | average angle error: 36.3 -> 27.7 | Improved |
| replace Wohlhart triplet/pair loss with pose-guided loss, single NN | pose descriptor · LINEMOD 5 degree accuracy | pose accuracy (%): 54% to 72% | Improved |
Closest papers by meaning
- Learning Local RGB-to-CAD Correspondences for Object Pose Estimation (ICCV 2019 · similarity 0.89)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.88)
- GIVEPose: Gradual Intra-class Variation Elimination for RGB-based Category-Level Object Pose Estimation (CVPR 2025 · similarity 0.88)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (ICLR 2024 · similarity 0.87)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.87)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.87)
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.
