ACDC: The Adverse Conditions Dataset with Correspondences for Robust Semantic Driving Scene Perception

The introduction of the ACDC dataset marks a significant advancement in the field of autonomous driving, particularly for Level-5 automation, which demands robust visual perception systems capable of interpreting images in various adverse conditions. Comprising 8012 images, ACDC includes 4006 images captured under challenging scenarios such as fog, nighttime, rain, and snow, each accompanied by high-quality pixel-level panoptic annotations. This comprehensive dataset not only addresses the shortcomings of existing datasets that predominantly feature normal conditions but also supports essential tasks like semantic segmentation, object detection, instance segmentation, and panoptic segmentation. The empirical studies conducted demonstrate the dataset's value in highlighting the challenges posed by adverse conditions to current state-of-the-art approaches, steering future research and development efforts in creating safer and more reliable autonomous driving technologies.