Automatically augmented and annotated urban datasets using mixed reality
Daniel Gómez Casañ · Dipòsit Digital de Documents de la UAB (Universitat Autònoma de Barcelona) · 2018
This project addresses the topic of database augmentation through a focus on process automation. In the field of autonomous driving systems the datasets used by the learning algorithms are decisive. Furthermore, the underlying machine learning systems would always benefit from having further quality data to learn from. A recent work on the topic of image dataset augmentation has been published, but it did not focus on the automation of the process, and it only involved the addition of cars to the existing images. On the other hand, our project has been developed to also support other kinds of objects. Moreover, our work has centered on developing an automatic pipeline that enables a continual augmentation of the dataset. Thanks to the efforts invested into the analysis of the source images and the automated rendering of virtual objects we can now produce augmented versions of the source images with relative ease.