AVOID: Dataset for Driving Scene Understanding in Adverse Conditions

This paper introduces a new dataset, AVOID, specifically designed to address the challenges of road scene understanding for self-driving cars under adverse visual conditions. The dataset's focus on unexpected road obstacles and its inclusion of various data modalities (semantic maps, depth maps, LiDAR data) make it valuable for training and evaluating perception models in realistic and challenging scenarios. The benchmarking and ablation studies further contribute to the paper's significance by providing insights into the performance of existing and proposed models.