H-GAN: the power of GANs in your Hands
This addresses the domain gap problem for computer vision researchers working with hand datasets, providing a method to leverage synthetic annotations in real-world applications, though it appears incremental as an adaptation of existing GAN techniques.
The paper tackles the problem of translating synthetic hand images to look realistic while preserving synthetic annotations, using a cycle-consistent adversarial learning approach with multi-scale perceptual discriminators. The result improves previous works in both qualitative and quantitative evaluations, with generated hands shown to be statistically similar to real hands in a classification task.
We present HandGAN (H-GAN), a cycle-consistent adversarial learning approach implementing multi-scale perceptual discriminators. It is designed to translate synthetic images of hands to the real domain. Synthetic hands provide complete ground-truth annotations, yet they are not representative of the target distribution of real-world data. We strive to provide the perfect blend of a realistic hand appearance with synthetic annotations. Relying on image-to-image translation, we improve the appearance of synthetic hands to approximate the statistical distribution underlying a collection of real images of hands. H-GAN tackles not only the cross-domain tone mapping but also structural differences in localized areas such as shading discontinuities. Results are evaluated on a qualitative and quantitative basis improving previous works. Furthermore, we relied on the hand classification task to claim our generated hands are statistically similar to the real domain of hands.