2018/11/26 by Girish Varma, Anbumani Subramanian, Varma, Girish +11 · 6 citations
Computer Science · Engineering · #Advanced Neural Network Applications #Autonomous Vehicle Technology and Safety #Video Surveillance and Tracking Methods
paper · pdf · doi:10.48550/arxiv.1811.10200
While several datasets for autonomous navigation have become available in\nrecent years, they tend to focus on structured driving environments. This\nusually corresponds to well-delineated infrastructure such as lanes, a small\nnumber of well-defined categories for traffic participants, low variation in\nobject or background appearance and strict adherence to traffic rules. We\npropose IDD, a novel dataset for road scene understanding in unstructured\nenvironments where the above assumptions are largely not satisfied. It consists\nof 10,004 images, finely annotated with 34 classes collected from 182 drive\nsequences on Indian roads. The label set is expanded in comparison to popular\nbenchmarks such as Cityscapes, to account for new classes. It also reflects\nlabel distributions of road scenes significantly different from existing\ndatasets, with most classes displaying greater within-class diversity.\nConsistent with real driving behaviours, it also identifies new classes such as\ndrivable areas besides the road. We propose a new four-level label hierarchy,\nwhich allows varying degrees of complexity and opens up possibilities for new\ntraining methods. Our empirical study provides an in-depth analysis of the\nlabel characteristics. State-of-the-art methods for semantic segmentation\nachieve much lower accuracies on our dataset, demonstrating its distinction\ncompared to Cityscapes. Finally, we propose that our dataset is an ideal\nopportunity for new problems such as domain adaptation, few-shot learning and\nbehaviour prediction in road scenes.\n