New generation autonomous vehicles are expected to successfully navigate in environments with challenging lighting conditions. This is where common, photometric visual perception-based tracking algorithms are put to their limits. We are now exploring the use of radiometric sensors, providing infrared frames (instead of photometric, RGB ones) for object tracking in low-light conditions. The challenge put forward by the Technology Innovation Institute's Autonomous Robotics Research Center is to develop approaches that can effectively track objects in the dark, in both structured and unstructured environments, including pedestrians, buggies, cars, and motorcyclists, among others.
The Infrared Tracking Challenge is open to innovators, start-ups, research institutes, and university students from anywhere in the world. Challengers are encouraged to leverage public datasets with similar content as TII's sample dataset and utilize color-to-thermal domain adaptation techniques in light of scarce publicly available annotated thermal video sequences. Submissions will be evaluated based on criteria set by TII, based on common tracking metrics in the RGB domain.
and final winner approvals for announcement
HeroX exists to enable anyone, anywhere in the world, to create a challenge that addresses any problem or opportunity, build a community around that challenge, and activate the circumstances that can lead to breakthrough innovation.
- Dockerized application of the challengers algorithm with clear installation instructions on an Ubuntu platform.
- The submission must accept a link to a ROS bagfile as input and output a .json file containing the predictions.
- Topics in the bagfile are same as the ones shared for training. Predictions are expected to be in same format as ground truth annotations.
- Each prediction must contain the following information: image frame number, 2D bounding box coordinates and ID of the instance.