Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised Secondary Landmark Detection via 3D Representation Learning
Praneet Bala, Jan Zimmermann, Hyun Soo Park, Benjamin Y. Hayden
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 multiview contrastive loss L_U^c (ablation) | L_L + L_U^g · Human3.6M | PCKh@0.5 mean: 0.43 -> 0.57 | Improved |
| add multiview contrastive loss L_U^c (ablation) | L_L + L_U^g · DeepFly3D | PCKh@0.5 mean: 0.42 -> 0.79 | Improved |
| add multiview contrastive loss L_U^c (ablation) | L_L + L_U^g · OpenMonkeyPose | PCKh@0.5 mean: 0.74 -> 0.75 | No clear effect |
| add 3D shared-representation self-supervision L_U^g (ablation) | supervised L_L · Human3.6M | PCKh@0.5 mean: 0.12 -> 0.43 | Improved |
| add 3D shared-representation self-supervision L_U^g (ablation) | supervised L_L · DeepFly3D | PCKh@0.5 mean: 0.31 -> 0.42 | Improved |
| add 3D shared-representation self-supervision L_U^g (ablation) | supervised L_L · OpenMonkeyPose | PCKh@0.5 mean: 0.72 -> 0.74 | Improved |
| add triangulation reprojection self-supervision L_U^t (ablation) | supervised L_L · Human3.6M | PCKh@0.5 mean: 0.12 -> 0.09 | Hurt |
| add triangulation reprojection self-supervision L_U^t (ablation) | supervised L_L · DeepFly3D | PCKh@0.5 mean: 0.31 -> 0.28 | Hurt |
| use 3D instead of 2D primary-landmark representation for predicting secondary landmarks (ablation) | linear subspace analysis · OpenMonkeyPose | secondary landmark reconstruction error: 30.2-60.6% error reduction | Improved |
| use semi-supervised multiview training instead of supervised only (2k labeled) (ablation) | Ours · OpenMonkeyPose (2k labeled) | PCKh secondary landmarks: 0.567 -> 0.413 | Hurt |
| use semi-supervised multiview training instead of supervised only (14k labeled) (ablation) | Ours · OpenMonkeyPose (14k labeled) | PCKh secondary landmarks: 0.722 -> 0.753 | Improved |
| remove wrist primary landmarks from the input configuration (ablation) | shared-representation predictor · OpenMonkeyPose | elbow reconstruction error | Hurt |
Closest papers by meaning
- MBW: Multi-view Bootstrapping in the Wild (NeurIPS 2022 · similarity 0.89)
- BigMaQ: A Big Macaque Motion and Animation Dataset Bridging Image and 3D Pose Representations (ICLR 2026 · similarity 0.87)
- HieDG: A Hierarchical Discrete Geometry-Guided Framework for Multi-Animal Tracking (ECCV 2026 · similarity 0.87)
- BEAST3D: animal behavioral analysis and neural encoding from multi-view video via Gaussian splatting (NeurIPS 2026 · similarity 0.86)
- Pose Splatter: A 3D Gaussian Splatting Model for Quantifying Animal Pose and Appearance (NeurIPS 2025 · similarity 0.86)
- FaceLift: Semi-supervised 3D Facial Landmark Localization (CVPR 2024 · similarity 0.84)
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.
