Research map / 3D vision / Human pose estimation
EgoPressure: A Dataset for Hand Pressure and Pose Estimation in Egocentric Vision
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces EgoPressure, an egocentric RGB-D dataset with pressure-sensor labels and multi-view optimized MANO hand meshes, plus baselines for image-projected and hand-projected (UV) pressure estimation including PressureFormer.
Problem
Estimating hand pressure and contact from egocentric RGB images is hard because no comprehensive datasets provide both hand pose and pressure annotations.
Approach
A multi-camera marker-less annotation method (HaMeR init, SAM masks, differentiable rendering with a virtual orthographic camera under the touchpad) that projects touchpad pressure onto a MANO hand mesh UV texture, and PressureFormer, which predicts pressure as a UV map from HaMeR vertices and ViT tokens.
Key results
- Adding HaMeR-predicted hand pose to PressureVisionNet raises egocentric volumetric IoU from 38.64% to 40.52%, and GT pose to 41.39%.
- PressureFormer reaches 33.12% UV pressure contact IoU and 24.54% UV volumetric IoU on EgoPressure (ego. & exo.), versus 21.53% and 16.41% for PressureVisionNet.
- Under cross-view generalization (train exo 2,3,4,5; test 1,6,7), HaMeR-pose augmentation raises volumetric IoU from 25.05% to 28.10%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add HaMeR-estimated hand pose | PressureVisionNet [19] · EgoPressure ego. test | Vol. IoU: 38.64 -> 40.52 | Helped |
| add ground-truth hand pose | PressureVisionNet [19] · EgoPressure ego. test | Vol. IoU: 38.64 -> 41.39 | Helped |
| add HaMeR-estimated hand pose | PressureVisionNet [19] · EgoPressure exo. views 2,3,4,5 test | Vol. IoU: 44.73 -> 45.01 | Helped |
| add ground-truth hand pose | PressureVisionNet [19] · EgoPressure exo. views 2,3,4,5 test | Vol. IoU: 44.73 -> 47.58 | Helped |
| add HaMeR-estimated hand pose | PressureVisionNet [19] · EgoPressure exo. views 1,6,7 test; trained on views 2,3,4,5 | Vol. IoU: 25.05 -> 28.10 | Helped |
| add ground-truth hand pose | PressureVisionNet [19] · EgoPressure exo. views 1,6,7 test; trained on views 2,3,4,5 | Vol. IoU: 25.05 -> 31.39 | Helped |
| predict pressure on the hand UV map with PressureFormer | PressureVisionNet [19] · EgoPressure ego. & exo. | UV Press. Contact IoU: 21.53 -> 33.12 | Helped |
| remove coarse UV-pressure loss Lc (ablation) | PressureFormer · EgoPressure ego. & exo. | UV Press. Contact IoU: 33.12 -> 26.24 | Hurt |
| remove coarse UV-pressure loss Lc (ablation) | PressureFormer · EgoPressure ego. & exo. | UV Press. Vol. IoU: 24.54 -> 18.61 | Hurt |
| use PressureFormer instead of PressureVisionNet | Pressure estimation model · PressureVision exo. | Im. Contact IoU: 7.54 -> 29.03 | Helped |
Surprising findings
- The hand-pose-augmented image-projected baseline achieves the best image volumetric IoU, rather than PressureFormer.
Limitations
- Dataset limited to hand interactions with flat surfaces.
- Collected indoors with only single-hand interactions.
- Participants are mostly from one institution (21 participants, ages 23-32).
Experimental setup
Datasets: EgoPressure PressureVision
Benchmarks: EgoPressure PressureVision test set
Metrics: contact IoU volumetric IoU MAE temporal accuracy UV pressure IoU
Baselines: PressureVisionNet PressureVisionNet with HaMeR pose PressureVisionNet with GT pose PressureFormer w/o UV loss
Builds on: PressureVision PressureVisionNet HaMeR HARP MANO
Closest papers by meaning
- HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos (CVPR 2025 · similarity 0.87)
- EgoTac: In-the-wild Tactile Prediction from Egocentric Vision (NeurIPS 2026 · similarity 0.87)
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · similarity 0.87)
- Estimating Ego-Body Pose from Doubly Sparse Egocentric Video Data (NeurIPS 2024 · similarity 0.87)
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.86)
- Estimating Body and Hand Motion in an Ego-sensed World (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.