Research map / 3D vision / Human pose estimation
Click Here: Human-Localized Keypoints as Guidance for Viewpoint Estimation
Ryan Szeto, Jason J. Corso
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 keypoint map and class conditioned learned attention (full CH-CNN) over fine-tuned image-only R4CNN | R4CNN fine-tuned · PASCAL 3D+ val (mean) | Acc pi/6 (%): 85.7 -> 90.7 | Improved |
| same, MedErr | R4CNN fine-tuned · PASCAL 3D+ val (mean) | MedErr (degrees): 6.74 -> 6.35 | Improved |
| add keypoint class to the keypoint map for the attention weights (ablation) | CH-CNN (keypoint map only) · PASCAL 3D+ val (mean) | Acc pi/6 (%): 85.6 -> 90.7 | Improved |
| add keypoint map to the keypoint class for the attention weights (ablation) | CH-CNN (keypoint class only) · PASCAL 3D+ val (mean) | Acc pi/6 (%): 86.8 -> 90.7 | Improved |
| learn attention from keypoint data instead of fixed Gaussian weight map (ablation) | Keypoint features · PASCAL 3D+ val (mean) | Acc pi/6 (%): 84.4 -> 90.7 | Improved |
| learn attention from keypoint data instead of fixed uniform weight map (ablation) | Keypoint features · PASCAL 3D+ val (mean) | Acc pi/6 (%): 85.4 -> 90.7 | Improved |
| provide the true keypoint map and class instead of blank ones at test time (ablation) | CH-CNN · PASCAL 3D+ val (mean) | Acc pi/6 (%): 74.1 -> 90.7 | Improved |
| provide the true keypoint class in addition to true map at test time (ablation) | CH-CNN · PASCAL 3D+ val (mean) | Acc pi/6 (%): 83.1 -> 90.7 | Improved |
| fine-tune R4CNN on the new synthetic keypoint-dataset renders | R4CNN · PASCAL 3D+ val (mean) | Acc pi/6 (%): 84.1 -> 85.7 | Improved |
| use Chebyshev distance transform keypoint map instead of Gaussian map (ablation) | CH-CNN · PASCAL 3D+ val (mean) | Acc pi/6 (%): 84.4 -> 90.7 | Improved |
| use Chebyshev instead of Euclidean distance transform keypoint map (ablation) | CH-CNN · PASCAL 3D+ val (mean) | MedErr (degrees): 6.31 -> 6.35 | No clear effect |
| use Chebyshev instead of Manhattan distance transform keypoint map (ablation) | CH-CNN · PASCAL 3D+ val (mean) | Acc pi/6 (%): 89.8 -> 90.7 | No clear effect |
Closest papers by meaning
- Neural Interactive Keypoint Detection (ICCV 2023 · similarity 0.88)
- KeyPose: Multi-View 3D Labeling and Keypoint Estimation for Transparent Objects (CVPR 2020 · similarity 0.87)
- KeypointNet: A Large-Scale 3D Keypoint Dataset Aggregated From Numerous Human Annotations (CVPR 2020 · similarity 0.87)
- Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning (NeurIPS 2018 · similarity 0.87)
- StarMap for Category-Agnostic Keypoint and Viewpoint Estimation (ECCV 2018 · similarity 0.86)
- MV-SSM: Multi-View State Space Modeling for 3D Human Pose Estimation (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.
