Research map / 3D vision / Human pose estimation
On Boosting Single-Frame 3D Human Pose Estimation via Monocular Videos
Zhi Li, Xuan Wang, Fei Wang, Peilin Jiang
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 |
|---|---|---|---|
| fine-tune on predictions refined by low-rank trajectory optimisation instead of raw predictions (ablation) | fine-tuning with plain loss · H36M validation (subject-wise protocol S1) | MPJPE: 94.5 -> 91.0 | Improved |
| add weighted fidelity loss to raw-prediction fine-tuning (ablation) | original supervision · H36M validation (subject-wise protocol S1) | MPJPE: 94.5 -> 93.0 | Improved |
| add weighted fidelity loss to optimised-prediction fine-tuning (ablation) | optimised supervision · H36M validation (subject-wise protocol S1) | MPJPE: 91.0 -> 90.5 | Improved |
| add bone-length consistency loss to raw-prediction fine-tuning (ablation) | original supervision + weighted fidelity · H36M validation (subject-wise protocol S1) | MPJPE: 93.0 -> 92.6 | Improved |
| add bone-length consistency loss to optimised fine-tuning (ablation) | optimised supervision + weighted fidelity · H36M validation (subject-wise protocol S1) | MPJPE: 90.5 -> 88.8 | Improved |
| full two-stage fine-tuning with unlabelled video | baseline trained on S1 · H36M validation (subject-wise protocol S1) | MPJPE: 97.7 -> 88.8 | Improved |
| trajectory optimisation of predictions on unlabelled H36M video (ablation) | H36M S5-8 predictions · H36M S5-8 trajectory optimisation (baseline trained on S1) | MPJPE: 91.20 -> 81.67 | Improved |
| trajectory optimisation of predictions on unlabelled 3DHP video (ablation) | 3DHP S2-8 predictions · 3DHP S2-8 optimisation (S1) | MPJPE: 132.25 -> 125.97 | Improved |
| semi-supervised training of the 2D-input baseline without ground truth 2D annotations | Pavllo semi-supervised · H36M validation (subject-wise protocol S1) | MPJPE: 98.6 -> 119.3 | Hurt |
| fine-tune on unlabelled videos of new hard actions | action-wise protocol hard · H36M validation, 4 hard unlabelled actions (action-wise protocol hard) | MPJPE: 138.09 -> 119.76 | Improved |
| fine-tune on unlabelled videos of new actions | action-wise protocol half · H36M validation, unlabelled half of actions | MPJPE: 98.92 -> 88.75 | Improved |
| fine-tune on unlabelled phoning video when labelled data is only discussion (class mismatch) | discussion-trained model · H36M Phoning | MPJPE: 144.5 -> 134.2 | Improved |
Closest papers by meaning
- Learning Monocular 3D Human Pose Estimation From Multi-View Images (CVPR 2018 · similarity 0.92)
- A Comparative Assessment of Accuracy in Video-Based Monocular Human Pose Estimation Frameworks (TPAMI 2026 · similarity 0.91)
- CanonPose: Self-Supervised Monocular 3D Human Pose Estimation in the Wild (CVPR 2021 · similarity 0.91)
- Monocular 3D Pose Estimation via Pose Grammar and Data Augmentation (TPAMI 2021 · similarity 0.90)
- Capturing the Motion of Every Joint: 3D Human Pose and Shape Estimation with Independent Tokens (ICLR 2023 · similarity 0.90)
- PoseMamba: Monocular 3D Human Pose Estimation with Bidirectional Global-Local Spatio-Temporal State Space Model (AAAI 2025 · similarity 0.89)
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.
