Research map / 3D vision / Human pose estimation
DiffusionPose: Markov-Optimized Diffusion Model for Human Pose Estimation
Zhigang Wang, Zhenguang Liu, Shaojing Fan, Sifan Wu, Yingying Jiao
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 Spatio-Temporal Feature Encoder (Mamba plus Transformer) to the ViT plus U-Net baseline (ablation) | ViT + U-Net baseline · PoseTrack2017 val | mAP: 81.2 -> 85.7 | Improved |
| replace U-Net with the Markov-based Pose Denoising Decoder (ablation) | ViT baseline · PoseTrack2017 val | mAP: 81.2 -> 86.1 | Improved |
| use both Spatio-Temporal Feature Encoder and Pose Denoising Decoder together (ablation) | DiffusionPose · PoseTrack2017 val | mAP: 81.2 -> 87.8 | Improved |
| use Spatio-Temporal Block (Mamba plus attention) instead of vanilla Transformer block (ablation) | Spatio-Temporal Feature Encoder · PoseTrack2017 val | mAP: 86.3 -> 87.8 | Improved |
| add vanilla Transformer block to encoder over no encoder (ablation) | DiffusionPose · PoseTrack2017 val | mAP: 86.1 -> 86.3 | No clear effect |
| add feature conditions Cf with cross-attention in the decoder (ablation) | Pose Denoising Decoder · PoseTrack2017 val | mAP: 85.7 -> 86.0 | Improved |
| add position conditions Cp with location-guide self-attention (ablation) | Pose Denoising Decoder · PoseTrack2017 val | mAP: 86.0 -> 86.5 | Improved |
| add Markov Optimization Layer (MRF message passing between joints) (ablation) | Pose Denoising Decoder · PoseTrack2017 val | mAP: 86.5 -> 87.2 | Improved |
| add Markov loss on limb length and angle (ablation) | DiffusionPose · PoseTrack2017 val | mAP: 87.2 -> 87.8 | Improved |
| use DiffusionPose instead of DiffPose on occlusion and fast-motion videos | DiffusionPose · Challenging-PoseTrack | mAP: 55.3 -> 59.7 | Improved |
| use single-step DDIM denoising at inference with DiffusionPose | DiffusionPose · PoseTrack2017 val | convergence and accuracy: converges within 10 epochs vs 30 for DiffPose | Improved |
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.
- DiffPose — DiffPose: Toward More Reliable 3D Pose Estimation (CVPR 2023) · better on 5 of 5 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
- Disentangled Diffusion-Based 3D Human Pose Estimation with Hierarchical Spatial and Temporal Denoiser (AAAI 2024 · similarity 0.91)
- $\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation (NeurIPS 2024 · similarity 0.90)
- DiffusionRegPose: Enhancing Multi-Person Pose Estimation using a Diffusion-Based End-to-End Regression Approach (CVPR 2024 · similarity 0.89)
- Object Pose Estimation via the Aggregation of Diffusion Features (CVPR 2024 · similarity 0.89)
- GRPose: Learning Graph Relations for Human Image Generation with Pose Priors (AAAI 2025 · similarity 0.88)
- Dual-Diffusion for Binocular 3D Human Pose Estimation (NeurIPS 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.
