Research map / 3D vision / Human pose estimation
Generic 3D Representation via Pose Estimation and Matching
Amir R. Zamir, Tilman Wekel, Pulkit Agrawal, Colin Wei, Jitendra Malik, Silvio Savarese
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 |
|---|---|---|---|
| use Euler angles instead of quaternions (ablation) | pose ConvNet · Street view test set (pose estimation) | median angular error (deg): 29.8 -> 17.7 | Improved |
| use robust loss instead of l2 loss (ablation) | pose ConvNet · Street view test set (pose estimation) | median angular error (deg): 22.3 -> 17.7 | Improved |
| train jointly on matching and pose (ablation) | joint descriptor · Street view test sets | matching and pose accuracy | No clear effect |
| use gradient clipping (ablation) | pose ConvNet · training | training stability | Improved |
| use our representation instead of AlexNet Pool5 for NN layout classification | frozen representation · LSUN layout classification | accuracy (%): 47.7% -> 57.6% | Improved |
| use our frozen representation instead of training from scratch | PASCAL3D pose · PASCAL3D pose estimation | average pose error (deg): 34 -> 26 | Improved |
| use our representation instead of ImageNet AlexNet | PASCAL3D pose · PASCAL3D pose estimation | average pose error (deg): 23 -> 26 | Hurt |
| use our descriptor instead of siam on Brown benchmark | descriptor matching · Brown benchmark (mean) | FPR@95 (%): 10.07 -> 8.46 | Improved |
| use our descriptor on small-baseline Lib to Yos transfer | descriptor matching · Brown benchmark (train Lib, test Yos) | FPR@95 (%): 11.18 -> 13.78 | Hurt |
| use our unsupervised representation vs supervised Hedau et al. for layout | frozen representation · LSUN layout estimation | corner error: 0.15 -> 0.16 | Hurt |
Closest papers by meaning
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.89)
- Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose (NeurIPS 2021 · similarity 0.89)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.88)
- Templates for 3D Object Pose Estimation Revisited: Generalization to New Objects and Robustness to Occlusions (CVPR 2022 · similarity 0.87)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (ICLR 2024 · similarity 0.86)
- FreeMan: Towards Benchmarking 3D Human Pose Estimation under Real-World Conditions (CVPR 2024 · similarity 0.86)
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.
