VoloGAN: Adversarial Domain Adaptation for Synthetic Depth Data

TL;DR

VoloGAN is a novel adversarial domain adaptation network that converts synthetic RGB-D images into realistic sensor-like images, enhancing training data quality for 3D reconstruction with minimal real data.

Contribution

It introduces a CycleGAN-based framework with specialized loss functions for realistic synthetic-to-real depth image translation, improving data generation for 3D reconstruction.

Findings

Effective synthetic-to-real depth image translation demonstrated

Improved training of 3D reconstruction models with limited real data

CycleGAN architecture tailored for RGB-D domain adaptation

Abstract

We present VoloGAN, an adversarial domain adaptation network that translates synthetic RGB-D images of a high-quality 3D model of a person, into RGB-D images that could be generated with a consumer depth sensor. This system is especially useful to generate high amount training data for single-view 3D reconstruction algorithms replicating the real-world capture conditions, being able to imitate the style of different sensor types, for the same high-end 3D model database. The network uses a CycleGAN framework with a U-Net architecture for the generator and a discriminator inspired by SIV-GAN. We use different optimizers and learning rate schedules to train the generator and the discriminator. We further construct a loss function that considers image channels individually and, among other metrics, evaluates the structural similarity. We demonstrate that CycleGANs can be used to apply adversarial domain adaptation of synthetic 3D data to train a volumetric video generator model having only few training samples.