Computer Graphics Laboratory ETH Zurich


Monocular RGB Hand Pose Inference from Unsupervised Refinable Nets

E. Dibra, S. Melchior, A. Balkis, T. Wolf, C. Öztireli, M. Gross

IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops (3D Humans 2018) (Salt Lake City, USA, June 18-22, 2018), pp. 1188-1198


CNN-based approaches are typically data-hungry, and when the task to solve is monocular RGB hand pose inference, obtaining real labelled training data is very hard to obtain. To overcome this, in this work we propose a new, large, realistically rendered, available hand dataset and a neural network trained on it, with the ability to refine itself to real unlabeled RGB images, given unlabeled corresponding depth images. We benchmark and validate our method on available and captured datasets, demonstrating that we strongly compare and even outperform state-of-the-art methods on tasks varying from 3D pose estimation to hand gesture recognition.


Download Paper
Download Paper
[PDF suppl.]
Download Video
Download Paper