BDD100K is a large-scale driving video dataset designed to support heterogeneous multitask learning in autonomous driving applications. The dataset includes 100,000 videos with diverse scenes, weather conditions, and geographic locations, covering a wide range of driving scenarios. It provides extensive annotations for various tasks such as image tagging, lane detection, drivable area segmentation, road object detection, semantic segmentation, instance segmentation, multi-object detection tracking, multi-object segmentation tracking, domain adaptation, and imitation learning. The dataset aims to address the limitations of existing driving datasets by offering a more comprehensive and diverse set of visual content and annotations. Experiments conducted on BDD100K reveal the need for specialized training strategies to handle heterogeneous tasks, and the dataset serves as a benchmark for future research in multitask learning for autonomous driving.BDD100K is a large-scale driving video dataset designed to support heterogeneous multitask learning in autonomous driving applications. The dataset includes 100,000 videos with diverse scenes, weather conditions, and geographic locations, covering a wide range of driving scenarios. It provides extensive annotations for various tasks such as image tagging, lane detection, drivable area segmentation, road object detection, semantic segmentation, instance segmentation, multi-object detection tracking, multi-object segmentation tracking, domain adaptation, and imitation learning. The dataset aims to address the limitations of existing driving datasets by offering a more comprehensive and diverse set of visual content and annotations. Experiments conducted on BDD100K reveal the need for specialized training strategies to handle heterogeneous tasks, and the dataset serves as a benchmark for future research in multitask learning for autonomous driving.