IIIT-H Expands Indian Driving Dataset For Research On India’s Chaotic Roads
The dataset now has over 46,000 annotated images and more than 12,000 LiDAR frames, and is being used by researchers in 30 countries.
Hyderabad:IIIT-Hyderabad has expanded the Indian Driving Dataset (IDD), first launched in 2018, into one of the most widely used resources on unstructured traffic conditions. The dataset now has over 46,000 annotated images and more than 12,000 LiDAR frames, and is being used by researchers in 30 countries.
The dataset, created with support from Intel Bangalore and UC San Diego, was built with 10,004 images captured in Hyderabad and Bengaluru. It was designed to reflect categories missing in Western collections such as autorickshaws, cattle, and undefined road edges.
Prof. C.V. Jawahar, who led the project, said the aim was to build something that mirrored Indian reality. “Most smart vehicles are trained on Western roads. We wanted to objectively show how our roads are different and create data that could serve Indian conditions,” Prof. Jawahar told Deccan Chronicle.
Over time, the dataset has grown into multiple formats including IDD-Detection, IDD-3D, and smaller “lite” versions for students and smaller labs. IIIT-H also developed the Bodhyaan research vehicle, fitted with cameras and sensors to collect multimodal driving data.
Dr Anbumani Subramanian, adjunct faculty at IIIT-H, said the dataset has reached a wide global audience. “We now have over 15,000 downloads with 10,000 users across 88 countries. It has helped students, startups and researchers build new algorithms and test them in real-world conditions,” he said.
The research has gone beyond self-driving cars. “The same object detection tools can be applied to pothole detection, tree counting, and even estimating flood damage after heavy rains,” Prof. Jawahar added. Telangana officials had earlier approached the team to adapt their algorithms for measuring Hyderabad’s green cover.
A recent addition is a two-wheeler dataset. “Since 75 per cent of India’s vehicles are motorbikes, we needed to study mobility from their perspective too. This has applications in EV range, fall detection and rider safety,” Dr Subramanian explained.
