Sim2Real for Self-Supervised Monocular Depth and Segmentation

TL;DR

This paper introduces a twin VAE-based domain adaptation method that enables training perception models for depth and segmentation in simulation, which then effectively transfer to real-world autonomous vehicle data without requiring real ground-truth labels.

Contribution

The paper presents a novel twin VAE architecture with shared latent space for sim2real transfer, eliminating the need for real domain labels during training.

Findings

Effective sim2real transfer of depth and segmentation tasks

Comparable performance to supervised methods in real domain

No paired real-world ground-truth data needed for training

Abstract

Image-based learning methods for autonomous vehicle perception tasks require large quantities of labelled, real data in order to properly train without overfitting, which can often be incredibly costly. While leveraging the power of simulated data can potentially aid in mitigating these costs, networks trained in the simulation domain usually fail to perform adequately when applied to images in the real domain. Recent advances in domain adaptation have indicated that a shared latent space assumption can help to bridge the gap between the simulation and real domains, allowing the transference of the predictive capabilities of a network from the simulation domain to the real domain. We demonstrate that a twin VAE-based architecture with a shared latent space and auxiliary decoders is able to bridge the sim2real gap without requiring any paired, ground-truth data in the real domain. Using only paired, ground-truth data in the simulation domain, this architecture has the potential to generate perception tasks such as depth and segmentation maps. We compare this method to networks trained in a supervised manner to indicate the merit of these results.