Logo image
Back
Domain-Transferred Synthetic Data Generation for Improving Monocular Depth Estimation
Conference proceeding

Domain-Transferred Synthetic Data Generation for Improving Monocular Depth Estimation

Knut Peterson, Seungyeop Lee, Solmaz Arezoomandan and David Han
International Conference on Control, Automation and Systems (Online), pp 203-208
04 Nov 2025
DOI: https://doi.org/10.23919/ICCAS66577.2025.11301203

Abstract

Data collection Data models Depth measurement Monocular Depth Estimation Sensors Sim-to-Real Transfer Solid modeling Synthetic data Three-dimensional displays Training Computer Vision Machine Learning Noise
A major obstacle to the development of effective monocular depth estimation algorithms is the difficulty in obtaining high-quality metric depth data that corresponds to real-world RGB images. Collecting this data is timeconsuming and costly, and even data collected by modern sensors has limited range or resolution, and is subject to inconsistencies and noise. Data generated in simulation avoids these problems with accurate depth information, but models trained on synthetic data often do not transfer well to real world applications. To combat this, we propose a method of data generation in simulation using 3D synthetic environments and CycleGAN domain transfer to increase the realism of simulated images. We analyze this data generation method by training multiple depth estimation models on different datasets, including synthetic and domain-transferred data. We evaluate the performance of the models on the NYU-Depth V2 dataset to verify the generalizability of the approach and show that GAN-transformed data effectively helps to bridge the gap between simulated and real-world data in depth estimation.

Metrics

9 Record Views

Details

Logo image