Oxford Road Boundaries Dataset

Welcome to the Oxford Road Boundaries dataset, designed for training and testing machine-learning-based road-boundary detection and inference approaches. We have hand-annotated two of the 10km long forays from the Oxford Robotcar Dataset and subsequently generated thousands of images with semi-annotated road boundary masks. To boost the number of training samples, we used a vision-based localiser to project labels from the annotated datasets to other traversals at different times of data and weather conditions. The Oxford Road Boundaries dataset contains 62605 labelled samples, of which 47639 samples are curated. Each of these samples contain both raw and classified masks for left and right lenses. Our data contains images from a diverse set of scenarios such as straight roads, parked cars, and junctions.

Summary of data released

Foray # of frames (A) # of frames (C) % of occluded labels (A) % of occluded labels (C) Download size (A) Download size (C) Annotation
2015-05-26-13-59-22 22775 14890 31% 29% 63 GB 41 GB manual
2015-03-17-11-08-44 4869 3923 30% 30% 14 GB 10 GB automatic
2015-05-08-10-33-09 5519 4617 31% 31% 16 GB 13 GB automatic
2015-05-19-14-06-38 5424 4271 45% 44% 15 GB 12 GB automatic
2018-04-30-14-43-54 15153 13353 28% 29% 44 GB 38 GB manual
2019-01-10-11-46-21 4440 3392 39% 39% 12 GB 10 GB automatic
2019-01-10-12-32-52 4425 3193 40% 38% 12 GB 8 GB automatic
Total 62605 47639 33% 32% 176 GB 132 GB NA

A - all frames, C - curated frames