Research map / 3D vision / Human pose estimation
MixSynthFormer: A Transformer Encoder-like Structure with Mixed Synthetic Self-attention for Efficient Human Pose Estimation
Yuran Sun, Alan William Dougherty, Zhuoying Zhang, Yi King Choi, Chuan Wu
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 |
|---|---|---|---|
| Remove spatial attention from MixSynth Attention (ablation) | MixSynthFormer · Sub-JHMDB | PCK@0.1 (%): 92.9 -> 95.1 | Improved |
| Add SE layer to attention matrix generation (ablation) | MixSynthFormer · Sub-JHMDB | PCK@0.1 (%): 94.2 -> 95.1 | Improved |
| Add weighted-sum combination of spatial and temporal branches (ablation) | MixSynthFormer · Sub-JHMDB | PCK@0.1 (%): 94.7 -> 95.1 | Improved |
| Add temporal attention (ablation) | MixSynthFormer · Sub-JHMDB | PCK@0.1 (%): 94.4 -> 95.1 | Improved |
| Use synthetic attention instead of vanilla self-attention (ablation) | Temporal attention · 3DPW | MPJPE: 92.6 -> 92.1; params 0.14M -> 0.10M | Improved |
| Apply reduction factor 4 in temporal attention synthesis (ablation) | MixSynthFormer · Human3.6M | MPJPE: 52.2 -> 51.4 | Improved |
| Apply reduction factor on both spatial and temporal dimensions (ablation) | MixSynthFormer · Human3.6M | MPJPE: 52.2 -> 51.8; params 0.26M -> 0.18M | Mixed |
| Use MixSynthFormer instead of DeciWatch | Sampling-based pose estimation · Human3.6M | MPJPE: 52.8 -> 50.9 | Improved |
| Use MixSynthFormer instead of DeciWatch on AIST++ | Sampling-based pose estimation · AIST++ | Accel error: 7.1 -> 4.7 | Improved |
| Use MixSynthFormer for longer prediction horizons | Motion prediction · Human3.6M 400 ms | MPJPE: 80 ms: 10.1 -> 8.4; 400 ms: 38.3 -> 37.4 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeciWatch — DeciWatch: A Simple Baseline for 10× Efficient 2D and 3D Pose Estimation (ECCV 2022) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video (CVPR 2022 · similarity 0.89)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.88)
- PoseFormerV2: Exploring Frequency Domain for Efficient and Robust 3D Human Pose Estimation (CVPR 2023 · similarity 0.88)
- SVTformer: Spatial-View-Temporal Transformer for Multi-View 3D Human Pose Estimation (AAAI 2025 · similarity 0.87)
- PoseCrafter: Extreme Pose Estimation with Hybrid Video Synthesis (NeurIPS 2025 · similarity 0.86)
- DiffPerformer: Iterative Learning of Consistent Latent Guidance for Diffusion-based Human Video Generation (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.
