Research map / 3D vision / Human pose estimation
Active Learning for Human Pose Estimation
Buyu Liu, Vittorio Ferrari
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 Multiple Peak Entropy instead of Highest Probability / Best-vs-Second-Best uncertainty (ablation) | image-level active learning · MPII validation | PCK-h accuracy vs annotation fraction | Improved |
| select by active cues instead of Random (ablation) | active learning · MPII validation | PCK-h accuracy vs annotation fraction | Improved |
| use Dynamic Combination of influence and uncertainty instead of Static Combination (W fixed at 0.5) (ablation) | Inf + MPE · MPII validation | PCK-h accuracy vs annotation fraction | Improved |
| add influence cue early and uncertainty later (time-varying weight) (ablation) | active selection · MPII validation | PCK-h accuracy | Improved |
| use DC instead of HP at 30% annotation budget (ablation) | active selection · MPII validation | PCK-h accuracy: 8% better | Improved |
| use DC instead of RM at 30% annotation budget (ablation) | active selection · MPII validation | PCK-h accuracy: 19% better | Improved |
| use DC with 90% of data annotated instead of HP/RM (ablation) | active selection · MPII validation | PCK-h accuracy | No clear effect |
| select individual joints (DC-jt) instead of whole images (DC-im) at low budget (ablation) | DC · MPII validation | PCK-h accuracy: half the annotation effort of RM-im for 40% of performance | Improved |
| apply DC-im active learning on a strong LSP-trained CPM, adding 10% MPII | CPM · LSP test | PCK-h accuracy: 84.3% -> 86.7% | Improved |
| add 30% MPII images chosen by DC-im | CPM · LSP test | PCK-h accuracy: 84.3% -> 88.9% | Improved |
| replace standard exact-click interface with VD-based region interface | annotation process · MPII annotation time | annotation time per joint: about 33% time saved | Improved |
| train on VD-based (peak-selected) annotations instead of exact ground truth | DC-im · MPII validation | PCK-h accuracy: within 3% gap | No clear effect |
Closest papers by meaning
- Deep Reinforcement Learning for Active Human Pose Estimation (AAAI 2020 · similarity 0.90)
- Human Pose Estimation Using Global and Local Normalization (ICCV 2017 · similarity 0.88)
- Meta Agent Teaming Active Learning for Pose Estimation (CVPR 2022 · similarity 0.88)
- Pose Prior Learner: Unsupervised Categorical Prior Learning for Pose Estimation (ICLR 2026 · similarity 0.88)
- FlexPose: Pose Distribution Adaptation with Limited Guidance (AAAI 2025 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 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.
