Smart Mobility

Advancing the state-of-AI research in the area of road safety to reduce accidents & fatalities in the country.

INAI’s goal in this domain is to advance state-of-AI research in the area of road safety that aligns with VISION ZERO goal of the government to reduce road accidents & fatalities in the country. Towards this goal, INAI put together the world’s first driving dataset (IDD) of Indian driving conditions.

While several datasets for autonomous navigation have become available in recent years, they have tended to focus on structured driving environments. This usually corresponds to well-delineated infrastructure such as lanes, a small number of well-defined categories for traffic participants, low variation in object or background appearance and strong adherence to traffic rules.

The driving conditions in India are quite diverse and the traffic behaviour is highly unstructured compared with rest of the world. These driving conditions pose unique challenges that are yet unsolved, for research in artificial intelligence (AI) and machine learning (ML) systems, and hence offer an immense opportunity for possible technical innovations in AI/ML systems for better road safety.

Datasets

IDD is a novel dataset for road scene understanding in unstructured environments. It consists of 10,000 images, finely annotated with 34 classes collected from 182 drive sequences on Indian roads. The label set is expanded in comparison to popular benchmarks such as Cityscapes, to account for new classes.

The dataset consists of images obtained from a front facing camera attached to a car driven in Hyderabad and Bengaluru. The images are mostly of 1080p resolution, but there are also some images with 720p and other resolutions. In addition to enabling researchers develop algorithms for the unique Indian conditions, this dataset also provides an opportunity for the global research community to investigate emerging AI concepts and benchmark their solutions.

IDD- Temporal

Temporally nearby frames (+/- 15 frames) from the IDD Segmentation data

IDD — Detection

40,000 images with bounding box annotations; released 2018

IDD — Lite

Subsampled version for IDD for use in resource constrained training/deployment, architecture search; 50MB in size with 7 classes

IDD — Multimodal

Primary, Secondary and Supplemental download packages containing (i) stereo images from front camera at 15 fps, (ii) GPS points at 15 Hz – latitude & longitude, and (iii) 16-channel LIDAR and (iv) OBD data

IDD- Segmentation

20,000 images and fine semantic segmentation annotation (14K Train, 2K Val, 4K Test) from 350 drive sequences

Events

AutoNUE 2019 IDD Challenge

While several datasets for autonomous navigation have become available in recent years, they have tended to focus on structured driving environments. IDD is a novel dataset for road scene understanding in unstructured environments. It consists of 20,000 images, finely annotated with 34 classes collected over 200 drive sequences on Indian roads. The label set is expanded in comparison to popular benchmarks such as Cityscapes, to account for new classes.

The challenge features:
• Datasets for segmentation, localization
• More than 20,000 annotated images hand picked from unstructured environments and occurrence of rare events
• More associated data including nearby frames, LIDAR data, GPS which can be used for specific challenges
• Challenges for resource constrained models with specific runtime budgets.

NCVPRIPG 2019 IDD Challenge

Following up on the AutoNUE 2019 IDD challenge, we launched a new challenge at NCVPRIPG 2019 IDD for students in Indian universities and colleges with the following enhancement.
• Semantic segmentation on IDD Lite Dataset

IDD-Lite, less than 50MB in size, is small and compact to fit on any personal computer and so will not need huge-compute infrastructure. It contains 7 classes (compared to 30 in IDD).

Publications

Proceedings of the 31st British Machine Vision Conference (BMVC), 2020

Spatial Feedback Learning to Improve Semantic Segmentation in Hot Weather

Shyam Nandan Rai, Vineeth N Balasubramanian, Anbumani Subramanian , C.V. Jawahar

Proceedings of the 31st British Machine Vision Conference (BMVC), 2020

Munich to Dubai: How far is it for Semantic Segmentation?

Shyam Nandan Rai, Vineeth N Balasubramanian, Anbumani Subramanian and C. V. Jawahar

International Conference on Robotics and Automation (ICRA), 2020

RoadText-1K: Text Detection & Recognition Dataset for Driving Videos

Sangeeth Reddy, Minesh Mathew, Lluis Gomez, Marçal Rusinol, Dimosthenis Karatzas and C. V. Jawahar

National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), 2019

Semantic Segmentation Datasets for Resource Constrained Training

Ashutosh Mishra, Sudhir Kumar, Tarun Kalluri, Girish Varma, Anbumani Subramaian, Manmohan Chandraker and C. V. Jawahar

National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), 2019

Dear Commissioner, please fix these: A scalable system for inspecting road infrastructure

Raghava Modhugu, Ranjith Reddy and C. V. Jawahar

International Conference on Computer Vision (ICCV), 2019

Universal Semi-Supervised Semantic Segmentation

Tarun Kalluri, Girish Varma, Manmohan Chandraker and C. V. Jawahar

IEEE Winter Conference on Applications of Computer Vision (WACV), 2019

IDD: A Dataset for Exploring Problems of Autonomous Navigation in Unconstrained Environments

Girish Varma, Anbumani Subramanian, Anoop Namboodiri, Manmohan Chandraker and C V Jawahar