Showcasing an ambitious algorithm for autonomous aerial navigation
Despite significant progress in recent years, most learning-based approaches to visual absolute localization, mapping or scene understanding target at a single domain training dataset and require a dense database of geotagged images to function well. To mitigate the data scarcity issue and improve the scalability of the neural models for autonomous aerial navigation, mapping/landscape digitizing and environmental monitoring, we developed TOPO-DataGen, a versatile synthetic data generation tool that traverses smoothly between the real and virtual world, hinged on the geographic camera viewpoint. We have used the designed workflow to generate a new first of a kind large-scale sim-to-real benchmark datasets in order to showcase and evaluate the utility of the said synthetic data. We have recently present (submitted a publication to CVPR 2022) for review of a visual localization system that learns to estimate camera poses in the real world with the help of synthetic data as a showcase of our open data generation tool and datasets (The CrossLoc datasets) created. Our experiments reveal that synthetic data generically enhances the neural network performance on real data. Furthermore, we introduce CrossLoc, a crossmodal visual representation learning approach to pose estimation that makes full use of the scene coordinate ground truth via self-supervision. Without any extra data, CrossLoc significantly outperforms all the state-of-the-art methods and achieves substantially higher real-data sample efficiency.
With the ENAC Open Software Support grant, researchers hope to get help sharing their work with scientific community and society in the form of an interactive project website. This website will display animations, share open training data as well as the source code and demos of the algorithms. ENAC-IT4R will also help the researcher to develop the contents of the website, in particular the publication pipeline of the model (implementation of code tests, CI/CD, packaging).
- PI: Iordan Doytchinov, TOPO
- Implementation: ENAC-IT4R
- Start date: March 2022