volograms/vologan

VoloGAN: Adversarial Domain Adaptation for Synthetic Depth Data

VoloGAN

Paper: https://arxiv.org/abs/2207.09204

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.

Architecture:

Our framework consists of four models: two generators and two discriminators. The generators translate an RGB-D image from one domain into the respective other domain. The discriminators predict, whether a generated RGB-D is real or fake. We incorporate four loss terms: adversarial loss, a channel-wise cycle-consistency loss, a channel-wise structural similarity loss of cycled image pairs and an identity loss.

Show Models

Generator	Discriminator
Our generator follows an encoder-decoder architecture with multiple connections between encoder and decoder.	The discriminator has three outputs to evaluate weather an input RGB-D image is real or fake: low level evaluation, layout evaluation and content evaluation. We explicitly encourage the disentanglement between layout and content by a two-branch architecture.

Results:

Generated RGBD

Generated RGB	Generated Depth

3D Point Clouds

Input RGBD (Rendered 3D Model)	Generated RGBD (Consumer Depth Sensor)
Input RGBD that has been rendered from an existing 3D model	Generated point cloud that incorporates the typical characteristics of a consumer depth sensor e.g. tail of points at edges.

PCA - Principal Component Analysis

Principal component analysis of the five principal components from 50 samples of each domain. Orange: generated images of the target domain. Blue: real images of the target domain.

Before Training	After Training