H-GAN: the power of GANs in your Hands

TL;DR

H-GAN is a novel cycle-consistent adversarial network that enhances synthetic hand images to closely resemble real images, improving realism and structural accuracy for better annotation transfer.

Contribution

It introduces multi-scale perceptual discriminators and a cycle-consistent framework for realistic hand image translation from synthetic data.

Findings

Improves realism of synthetic hand images

Achieves better statistical similarity to real data

Outperforms previous methods in qualitative and quantitative evaluations

Abstract

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.