Single-to-Dual-View Adaptation for Egocentric 3D Hand Pose Estimation

TL;DR

This paper introduces S2DHand, an unsupervised method that adapts single-view 3D hand pose estimators to dual-view settings without requiring multi-view annotations or fixed camera parameters, improving accuracy and versatility.

Contribution

The paper presents a novel unsupervised adaptation approach for dual-view hand pose estimation that works with arbitrary camera pairs and unknown parameters, unlike prior methods.

Findings

Significant accuracy improvements on various camera pairs

Effective cross-dataset generalization

Outperforms existing adaptation methods

Abstract

The pursuit of accurate 3D hand pose estimation stands as a keystone for understanding human activity in the realm of egocentric vision. The majority of existing estimation methods still rely on single-view images as input, leading to potential limitations, e.g., limited field-of-view and ambiguity in depth. To address these problems, adding another camera to better capture the shape of hands is a practical direction. However, existing multi-view hand pose estimation methods suffer from two main drawbacks: 1) Requiring multi-view annotations for training, which are expensive. 2) During testing, the model becomes inapplicable if camera parameters/layout are not the same as those used in training. In this paper, we propose a novel Single-to-Dual-view adaptation (S2DHand) solution that adapts a pre-trained single-view estimator to dual views. Compared with existing multi-view training methods, 1) our adaptation process is unsupervised, eliminating the need for multi-view annotation. 2) Moreover, our method can handle arbitrary dual-view pairs with unknown camera parameters, making the model applicable to diverse camera settings. Specifically, S2DHand is built on certain stereo constraints, including pair-wise cross-view consensus and invariance of transformation between both views. These two stereo constraints are used in a complementary manner to generate pseudo-labels, allowing reliable adaptation. Evaluation results reveal that S2DHand achieves significant improvements on arbitrary camera pairs under both in-dataset and cross-dataset settings, and outperforms existing adaptation methods with leading performance. Project page: https://github.com/MickeyLLG/S2DHand.