The EuroCity Persons dataset contains more than 47,000 photos of streets in 12 countries with annotations of pedestrians and vehicles. The dataset is aimed at improving object recognition models, including in occlusion conditions.
EuroCity Persons includes more than 200,000 annotations of pedestrian poses, which is about an order of magnitude more than in similar datasets. The table below shows the results of comparing EuroCity Persons with other datasets:
The data was obtained in 31 cities using cameras mounted on cars. The frame rate is 4 seconds. Each object in the images (bicycle, motorcycle, scooter, wheelchair, pedestrian and driver) is limited by a rectangular frame. Object annotations will include an approximate level of occlusion.
The shooting was carried out at all seasons of the year, at different times (day/night) and in different climatic conditions (including sunny and rainy weather, as well as during snowfall).
The dataset is available here.