MB7500 dataset contains a range of 7500 annoted images taken from a taken with a Phantom 4® drone, with an HD camera under windy conditions, which affected the image stabilizer capabilities.  Images were resized to 640 x 364 pixels, containing 41,040 ROI annotated objects, with a minimal height size set to 25 pixels. 60% of the annotated data corresponds to occluded motorcycles. Objects partially occluded with height less than 25 pixels were not annotated.

The dataset was annotated by means of Viper annotation tool. The ground truth generated is specified in a XML file which describes the class, frames covered by the object, Name, Id, height and and width of the bbox surrounding the object.


The Full HD entire Video sequence of 7500 photograms is available in a compressed file (VideoSequence(HD)). It corresponds to a AVi file with 1920 x 1080 pixels of resolution with MJPEG codec of 30 fps.

There is also a full sequece of 7500 photograms of 640 x 364 pixels (VideoSequence(Small))

If you want to access the photograms as a compressed folder click on ImagesSequence (zip), and the decompressed folder is available in ImagesSequence (dir)

The ground truth is a XML style file created by using Viper over the photograms sequence (GoundTruth).

Original Dataset Image vs Annotated Image.


We make the data available to the researchers in computer vision community, the only requirement for using MB7500 is to cite our paper:

title = {Motorcycle detection and classification in urban {Scenarios} using a model based on {Faster} {R}-{CNN}},
url = {},
abstract = {This paper introduces a Deep Learning Convolutional Neural Network model based on Faster-RCNN for motorcycle detection and classification on urban environments. The model is evaluated in occluded scenarios where more than 60\% of the vehicles present a degree of occlusion. For training and evaluation, we introduce a new dataset of 7500 annotated images, captured under real traffic scenes, using a drone mounted camera. Several tests were carried out to design the network, achieving promising results of 75\% in average precision (AP), even with the high number of occluded motorbikes, the low angle of capture and the moving camera. The model is also evaluated on low occlusions datasets, reaching results of up to 92\% in AP.},
urldate = {2018-08-13},
journal = {arXiv:1808.02299 [cs]},
author = {Espinosa, Jorge E. and Velastin, Sergio A. and Branch, John W.},
month = aug,
year = {2018},
note = {arXiv: 1808.02299},
keywords = {Computer Science - Computer Vision and Pattern Recognition},
annote = {Comment: Presented at 9th International Conference on Pattern Recognition Systems, ICPRS-18, 22-24 May 2018, Valparaiso, Chile},
file = {arXiv\:1808.02299 PDF:D\:\\zotero\\storage\\WKE9LZRQ\\Espinosa et al. - 2018 - Motorcycle detection and classification in urban S.pdf:application/pdf; Snapshot:D\:\\zotero\\storage\\8UIFNZNT\\1808.html:text/html}


On the table below, you can click on the links to download the data for the corresponding file format.

File Size
VideoSequence(HD) 2.11 GB
VideoSequence(Small) 411 MB
ImagesSequence (zip) 426MB
ImagesSequence (dir) 426MB
GoundTruth 3.3MB
Table 1 File Content


For any queries related to these datasets please email Jorge Espinosa