Research map / 3D vision / Human pose estimation
Motion Adaptive Pose Estimation From Compressed Videos
Zhipeng Fan, Jun Liu, Yao Wang
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 motion-vector warping of ConvLSTM hidden and cell states (ablation) | I-frame-only ConvLSTM · Penn Action (component ablation, ResNet34) | mean PCK: 93.8 -> 95.8 | Improved |
| add P-frame features at every frame to the warped ConvLSTM (ablation) | warped ConvLSTM · Penn Action (component ablation, ResNet34) | mean PCK: 95.8 -> 98.1 | Improved |
| add residual-driven dynamic gate instead of extracting features at every P-frame (ablation) | warped ConvLSTM with all P-frames · Penn Action (component ablation, ResNet34) | mean PCK: 98.1 -> 97.7 | No clear effect |
| drive the gate by residual frames instead of fully decoded P-frames (ablation) | gated ConvLSTM · Penn Action (component ablation, ResNet34) | mean PCK: 97.5 -> 97.7 | Improved |
| drive the gate by residual frames instead of fully decoded P-frames (ablation) | gated ConvLSTM · Penn Action (component ablation, ResNet34) | GFLOPs: 5.36 -> 4.10 | Improved |
| add temporal convolutions to the gate (ablation) | residual gate · Sub-JHMDB (gate design ablation) | keyframe percentage: 45.3% -> 38.7% | Improved |
| add temporal convolutions to the gate (ablation) | residual gate · Sub-JHMDB (gate design ablation) | mean PCK: 94.8 -> 94.6 | No clear effect |
| add sinusoidal positional encoding to the gate (ablation) | residual gate with temporal conv · Sub-JHMDB (gate design ablation) | keyframe percentage: 38.7% -> 35.2% | Improved |
| add sinusoidal positional encoding to the gate (ablation) | residual gate with temporal conv · Sub-JHMDB (gate design ablation) | mean PCK: 94.6 -> 94.7 | No clear effect |
| keep only 65% of frames as keyframes (ablation) | Motion Adaptive Pose Net · Penn Action | PCK: 0.001 drop | No clear effect |
| use ResNet34 instead of ResNet18 backbone (ablation) | Motion Adaptive Pose Net · Penn Action | mean PCK: 97.7 -> 97.7 | No clear effect |
| use motion-compressed pipeline vs KFP | KFP · Penn Action | mean PCK: 97.4 -> 97.7 | Improved |
Closest papers by meaning
- Learning Temporal Pose Estimation from Sparsely-Labeled Videos (NeurIPS 2019 · similarity 0.88)
- Direct Prediction of 3D Body Poses From Motion Compensated Sequences (CVPR 2016 · similarity 0.87)
- Deep Dual Consecutive Network for Human Pose Estimation (CVPR 2021 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.86)
- HumanMM: Global Human Motion Recovery from Multi-shot Videos (CVPR 2025 · similarity 0.86)
- Video-Based Human Pose Regression via Decoupled Space-Time Aggregation (CVPR 2024 · similarity 0.85)
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.
