Learning from Simulated and Unsupervised Images through Adversarial Training

TL;DR

This paper introduces a novel adversarial training method called Simulated+Unsupervised learning to enhance the realism of synthetic images, thereby improving model performance on real-world tasks without requiring labeled real data.

Contribution

The paper proposes a new adversarial training approach that preserves annotations while generating realistic images from synthetic data, reducing the synthetic-to-real gap.

Findings

Generated images are highly realistic, confirmed by qualitative analysis and user study.

Models trained on refined images outperform those trained on raw synthetic images.

Achieved state-of-the-art results on MPIIGaze dataset without using labeled real data.

Abstract

With recent progress in graphics, it has become more tractable to train models on synthetic images, potentially avoiding the need for expensive annotations. However, learning from synthetic images may not achieve the desired performance due to a gap between synthetic and real image distributions. To reduce this gap, we propose Simulated+Unsupervised (S+U) learning, where the task is to learn a model to improve the realism of a simulator's output using unlabeled real data, while preserving the annotation information from the simulator. We develop a method for S+U learning that uses an adversarial network similar to Generative Adversarial Networks (GANs), but with synthetic images as inputs instead of random vectors. We make several key modifications to the standard GAN algorithm to preserve annotations, avoid artifacts, and stabilize training: (i) a 'self-regularization' term, (ii) a local adversarial loss, and (iii) updating the discriminator using a history of refined images. We show that this enables generation of highly realistic images, which we demonstrate both qualitatively and with a user study. We quantitatively evaluate the generated images by training models for gaze estimation and hand pose estimation. We show a significant improvement over using synthetic images, and achieve state-of-the-art results on the MPIIGaze dataset without any labeled real data.