Research map / 3D vision / Human pose estimation
Camera Distortion-Aware 3D Human Pose Estimation in Video With Optimization-Based Meta-Learning
Hanbyel Cho, Yooshin Cho, Jaemyung Yu, Junmo 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 MAML with synthetic distorted tasks (ablation) | Pavllo et al. base model · Human3.6M distorted d1/d2, Scenario 1, 9 frames | MPJPE (mm): 84.2 / 79.6 -> 73.5 / 67.5 | Improved |
| add stratified sampling of distortion parameter (ablation) | MAML with synthetic tasks · Human3.6M distorted d1/d2, Scenario 1, 9 frames | MPJPE (mm): 73.5 / 67.5 -> 71.7 / 66.2 | Improved |
| add random distortion pretraining (ablation) | MAML with stratified sampling · Human3.6M distorted d1/d2, Scenario 1, 9 frames | MPJPE (mm): 71.7 / 66.2 -> 67.2 / 61.9 | Improved |
| add MAML with synthetic distorted tasks (ablation) | Pavllo et al. base model · Human3.6M distorted d1/d2, Scenario 1, 9 frames | PCKh@0.5 (%): 64.8 / 66.9 -> 72.0 / 74.5 | Improved |
| generate synthetic tasks from predicted 2D keypoints rather than ground-truth 3D joints (ablation) | Proposed method · Human3.6M distorted, Scenario 1, heavy | MPJPE (mm): 64.7 -> 62.0 | Improved |
| generate synthetic tasks from predicted 2D keypoints rather than ground-truth 3D joints (ablation) | Proposed method · Human3.6M distorted, Scenario 2, heavy | MPJPE (mm): 71.3 -> 66.1 | Improved |
| use proposed meta-learning method instead of baseline pose estimators | Pavllo et al. · Human3.6M distorted, Scenario 2, heavy | MPJPE (mm): 114.1 -> 66.1 | Improved |
| increase input frame length from 3 to 27 (ablation) | Proposed method · Human3.6M distorted, Scenario 1, all distortions | MPJPE (mm): 75.0 -> 57.6 | Improved |
| increase input frame length from 3 to 9 (ablation) | Proposed method · Human3.6M distorted, Scenario 1, all distortions | MPJPE (mm): 75.0 -> 59.8 | Improved |
| apply proposed method with 27 frames instead of Pavllo 27f | Pavllo et al. · Human3.6M distorted, Scenario 1, all distortions | MPJPE (mm): 72.4 -> 57.6 | Improved |
| pretrain on random distortion without meta-learning (ablation) | Random distortion pretraining · Human3.6M distorted | accuracy after adaptation | Hurt |
Closest papers by meaning
- Consensus-Based Optimization for 3D Human Pose Estimation in Camera Coordinates (IJCV 2022 · similarity 0.88)
- Robust 3D Human Pose Estimation from Single Images or Video Sequences (TPAMI 2018 · similarity 0.87)
- Revitalizing Optimization for 3D Human Pose and Shape Estimation: A Sparse Constrained Formulation (ICCV 2021 · similarity 0.87)
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.87)
- Kinematic-Structure-Preserved Representation for Unsupervised 3D Human Pose Estimation (AAAI 2020 · similarity 0.87)
- PoseIRM: Enhance 3D Human Pose Estimation on Unseen Camera Settings via Invariant Risk Minimization (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.
