Please refer to the following publications for the detailed descriptions to the datasets. C.Wang, Y.Fang, H.Zhao, C.Guo, S.Mita, H.Zha, Probabilistic Inference for Occluded and Multiview On-road Vehicle Detection, IEEE Trans. on Intelligent Transportation Systems, 17(1), 215-229, 2016. C.Wang, H.Zhao, C.Guo, S.Mita, H.Zha, On-road Vehicle Detection through Part Model Learning and Probabilistic Inference, IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 4965-4972, 2014. For questions, please email: zhaohj@cis.pku.edu.cn. --------------------------------------------------------------- For Ground Truth: In ground truth file, fno=23646 // frame number for correspoinding image 72 538 278 630 2 // bounding box for vehicle: left top x, left top y, right bottom x, right bottom, y, vehicle types (viewpoint class, or if occluded) viewpoint class: 1,5; refer to viewpoint class 1 in paper 2,6; refer to viewpoint class 2 3,7; refer to viewpoint class 3 4,8; refer to viewpoint class 4 9; occluded vehicle For occluded vehicles, we simply labeled partial observed vehicles, it's very subjective and very hard to apply strict rules for which occluded vehicle should be and could be detected. In our experiment we only consider the vehicle in current road. And we cut the test paranomic image for partial observation in image edge in experiment.