A Controlled Corruption Dataset for Autonomous Driving (DRIVE-C)

DRIVE-C is a controlled corruption dataset for autonomous driving, designed to support research in perception robustness, degradation awareness, and sensor health estimation. The dataset is built from real-world driving data and augmented with 12 types of physically inspired degradations across 5 severity levels. The current release includes:- 10 scenarios- 610 video clips (~78,000 frames)- Ground-truth and predicted Global Sensor Health Index (GSHI) This dataset enables controlled evaluation of perception models under realistic degradation conditions. Code and pipeline are available at:https://github.com/shiv-aher/drive-c-dataset