Research map / 3D vision / Human pose estimation
Temporal Representation Learning on Monocular Videos for 3D Human Pose Estimation
Sina Honari, Victor Constantin, Helge Rhodin, Mathieu Salzmann, Pascal Fua
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 decoder (image reconstruction) (ablation) | DSL-STN · H36M 0.3% labels | N-MPJPE: 236.32 -> 185.75 | Improved |
| add latent split into time-variant/invariant (ablation) | DSL-STN-Dec · H36M 0.3% labels | N-MPJPE: 185.75 -> 149.32 | Improved |
| add latent split (ablation) | DSL-STN-Dec · H36M all labels | N-MPJPE: 127.52 -> 95.39 | Improved |
| use DSL contrastive loss instead of CSS loss (ablation) | CSS-STN-Dec-Split · H36M 0.3% labels | N-MPJPE: 163.61 -> 149.32 | Improved |
| add contrastive loss over plain autoencoder (ablation) | AE-STN-Dec · H36M all labels | N-MPJPE: 114.73 -> 95.39 | Improved |
| add STN to DSL without decoding (ablation) | DSL · H36M 0.3% labels | N-MPJPE: 227.24 -> 236.32 | Hurt |
| add tracking loss (ablation) | Ours · Diving 5% labels | N-MPJPE: 296.22 -> 265.09 | Improved |
| use single-view model instead of DrNet | Ours · H36M 0.3% labels | N-MPJPE: 158.81 -> 149.32 | Improved |
| add 2D pose labels in feature extraction (ablation) | Resynthesis · Diving diver-split | N-MPJPE (cm): 15.7 -> 14.1 | Improved |
| replace ResNet50 with ResNet101 (ablation) | Resynthesis + 2D · 3DPW | N-MPJPE (cm): 13.29 -> 10.38 | Improved |
| use resynthesis pre-training instead of ImageNet pre-training (ablation) | ResNet50 + 2D · 3DPW | N-MPJPE (cm): 18.07 -> 13.29 | Improved |
| single-view Ours vs multi-view NSD | Ours · MPI-INF-3DHP 0.02% labels (MV-data) | N-MPJPE: 256.34 -> 261.99 | Hurt |
Closest papers by meaning
- On Boosting Single-Frame 3D Human Pose Estimation via Monocular Videos (ICCV 2019 · similarity 0.90)
- CanonPose: Self-Supervised Monocular 3D Human Pose Estimation in the Wild (CVPR 2021 · similarity 0.89)
- Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction (CVPR 2020 · similarity 0.89)
- Multiview-Consistent Semi-Supervised Learning for 3D Human Pose Estimation (CVPR 2020 · similarity 0.89)
- Non-local Latent Relation Distillation for Self-Adaptive 3D Human Pose Estimation (NeurIPS 2021 · similarity 0.88)
- Learning Monocular 3D Human Pose Estimation From Multi-View Images (CVPR 2018 · similarity 0.88)
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.
