AgriDrive: Data, Benchmarks and Analysis

With the rapid development of autonomous driving technology in the recent years, its application has been extended from urban driving to agricultural scenes. Obstacle detection and scene segmentation are the key parts of autonomous driving technology. Because of the difference between the agricultural and the urban scene, the existing algorithms of obstacle detection and feasible area segmentation may not be directly applied to the agricultural scene. In this paper, we construct a dataset of agricultural scene autonomous driving, which is used for obstacle detection and scene segmentation tasks. Besides, we evaluate the current state of the art deep learning models on this dataset. The result shows this dataset presents challenges for existing algorithms of obstacle detection and scene segmentation. Our dataset can be used for further research and algorithm development.