Scene understanding for applications such as intelligence, surveillance, and reconnaissance (ISR) and autonomous vehicles becomes extremely challenging in adverse weather conditions such as haze, fog and mist. These atmospheric phenomena with smoke particles and minute water droplets often result in imagery with non-linear noise, blur, reduced contrast levels and color dimming issues. Thus, the on-board vision systems are significantly obscured. These visual artifacts generated from uncontrolled and potentially changing environment or Degraded Visual Environments (DVE) poses major challenges in image enhancement and restoration, and object detection and classification, some of the key tasks towards the final goal of semantic scene understanding.
UG2+ Track 1 aims to evaluate and advance object detection algorithms’ robustness on images captured from hazy environmental situations. Participants are allowed to use a restoration/enhancement pre-processing step in the detection pipeline. In other words, they will not be tasked with the creation of novel object detection algorithms. A list of detection algorithms will be provided to them in order to facilitate studies of the interaction between image restoration and enhancement algorithms and the detectors. During the evaluation, the selected detection algorithms will be run on the test images. Through this challenge and benchmark, we aim to encourage more state-of-the-art single-image dehazing, haze quantification, and object detection algorithms.
Track 1 is based on the A2I2-Haze, the first real haze dataset with in-situ smoke measurement aligned to aerial imagery. A2I2-Haze has paired haze and haze-free imagery that will allow fine-grained evaluation of computer vision algorithms. A2I2-Haze is a result of joint collaboration with DEVCOM Army Research Laboratory, produced and measured by nonlethal smoke/obscurant munitions and generated hazy conditions in a controlled way. We provide a total of 229 paired hazy/clean frame images extracted from 12 videos. A2I2-Haze is labeled with civilian vehicles, and the participating teams will be allowed to use any extra labeled/unlabeled training data is allowed. There will be 197 paired images for training and 32 reserved for final testing.
For more details, pleas refer to our paper: A2I2-Haze.pdf
If you have any questions about this challenge track please feel free to email email@example.com