MC-hands-1M: A glove-wearing hand dataset for pose estimation

TL;DR

This paper introduces MC-hands-1M, a synthetic dataset for 3D glove-wearing hand pose estimation, demonstrating its effectiveness in improving model performance on both synthetic and real images.

Contribution

The creation of a large synthetic dataset specifically for glove-wearing hand pose estimation and its use in fine-tuning existing models for better accuracy.

Findings

Significant performance improvement on glove-wearing hand images

Effective use of synthetic data for specialized pose estimation

Potential for enhancing hand tracking in specialized scenarios

Abstract

Nowadays, the need for large amounts of carefully and complexly annotated data for the training of computer vision modules continues to grow. Furthermore, although the research community presents state of the art solutions to many problems, there exist special cases, such as the pose estimation and tracking of a glove-wearing hand, where the general approaches tend to be unable to provide an accurate solution or fail completely. In this work, we are presenting a synthetic dataset1 for 3D pose estimation of glove-wearing hands, in order to depict the value of data synthesis in computer vision. The dataset is used to fine-tune a public hand joint detection model, achieving significant performance in both synthetic and real images of glove-wearing hands.