Object Detection Under Uncontrolled Acquisition Environment and Scene Context Constraints Challenge: ICIP 2023

Image acquisition conditions can significantly affect high-level tasks in computer vision, such as object detection and segmentation, depth estimation, scene understanding, or object tracking for many vision applications (robotics, medical imaging, Etc.). Many works have proved the weak robustness of object detection models against distorted images. Usually, two approaches enable enhancing the models'  robustness by designing a model with a dedicated architecture or performing a data augmentation. In the proposed challenge, we focus on data augmentation that consists of training models with generated distorted images corresponding to real and widespread scenarios in natural environments. However, standard distorted datasets only provide global and unrealistic  distortions resulting from image acquisition (noise, compression, contrast changing), camera (motion and defocus blur), or atmospheric conditions (rain, fog). To overcome these limitations, we propose a new versatile database derived from the well-known MS-COCO database to which we applied local and global photo-realistic distortions. Distortions are generated by exploiting the depth information of the objects in the scene as well as their semantics.  These new local distortions are generated by considering the scene context. This allows to explore real scenarios ignored in conventional databases dedicated to various computer vision applications. This challenge will perform the first comprehensive benchmark of the impact of a wide range of distortions on the performance of current object detection methods.

Visit the Challenge website for details and more information!