Research map / 3D vision / Human pose estimation
E-3DPSM: A State Machine for Event-based Egocentric 3D Human Pose Estimation
Mayur Deshmukh, Hiroyasu Akada, Helge Rhodin, Christian Theobalt, Vladislav Golyanik
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 SSM blocks to the spatiotemporal encoder (ablation) | E-3DPSM · EE3D-R | MPJPE: 118.53 -> 84.45 | Improved |
| add SSM blocks at all stages instead of only stage four (ablation) | E-3DPSM · EE3D-R | MPJPE: 90.18 -> 84.45 | Improved |
| add deformable attention (ablation) | E-3DPSM · EE3D-R | MPJPE: 88.27 -> 84.45 | Improved |
| add learned Kalman-style fusion of delta and direct poses (ablation) | E-3DPSM · EE3D-R | MPJPE: 141.22 -> 84.45 | Improved |
| add delta-pose regression alongside direct pose (ablation) | E-3DPSM · EE3D-R | e_smooth: 17.22 -> 8.40 | Improved |
| make Kalman fusion learnable instead of static (ablation) | E-3DPSM · EE3D-R | MPJPE: 88.31 -> 84.45 | Improved |
| use globally learned Q and R instead of input/state-dependent (ablation) | E-3DPSM · EE3D-R | MPJPE: 91.15 -> 84.45 | Improved |
| train non-causally (bidirectional) instead of forward-only (ablation) | E-3DPSM · EE3D-R | MPJPE: 89.88 -> 84.45 | Improved |
| increase training sequence length from 20 to 40 poses (ablation) | E-3DPSM · EE3D-R | MPJPE: 86.25 -> 84.45 | Improved |
| reset SSM states every 40 frames at inference (ablation) | E-3DPSM · EE3D-R | MPJPE: 84.45 -> 104.56 | Hurt |
| reset fusion states every 40 frames at inference (ablation) | E-3DPSM · EE3D-R | e_smooth: 8.40 -> 21.50 | Hurt |
| use fixed LNES instead of learned voxel-based representation (ablation) | E-3DPSM · EE3D-R | MPJPE: 100.25 -> 84.45 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- EventEgo3D — EventEgo3D: 3D Human Motion Capture from Egocentric Event Streams (CVPR 2024) · better on 8 of 10 shared results
- EgoPoseFormer — EgoPoseFormer: A Simple Baseline for Stereo Egocentric 3D Human Pose Estimation (ECCV 2024) · better on 4 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
- EventEgo3D: 3D Human Motion Capture from Egocentric Event Streams (CVPR 2024 · similarity 0.89)
- EgoPoseFormer v2: Accurate Egocentric Human Motion Estimation for AR/VR (CVPR 2026 · similarity 0.88)
- Dynamics-regulated kinematic policy for egocentric pose estimation (NeurIPS 2021 · similarity 0.88)
- 3D Human Pose Perception from Egocentric Stereo Videos (CVPR 2024 · similarity 0.87)
- EgoSim: An Egocentric Multi-view Simulator and Real Dataset for Body-worn Cameras during Motion and Activity (NeurIPS 2024 · similarity 0.86)
- FRAME: Floor-aligned Representation for Avatar Motion from Egocentric Video (CVPR 2025 · similarity 0.86)
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.
