Sun Yat-sen University Detection And Tracking Of Moving Objects (DATMO) Dataset

Welcome to Sun Yat-sen University detection and tracking of moving objects dataset! This dataset is aiming at moving objects detection accuracy improvement. There are three packages named KITTI1, KITTI2 and SYSU in our DATMO dataset with manually labeled ground truth. KITTI1 and KITTI2 datasets are selected from KITTI stereo flow dataset and KITTI raw data while SYSU dataset are captured by our own car in daytime with resolution of 1299*374. There are 406 groups each contains current and previous stereo image pairs and there are 921 moving objects altogether. Moving objects contains various vehicles or pedestrians and there can upto 7 objects in a single image. The platform carrying cameras may be moving or static. Some samples are shown as follows:

The ground truth with coordinate of upper-left coner, width and height of boxs are also included in our dataset.

Relative DATMO dataset can be download here.

Sun Yat-sen University Night-time Vehicle Dataset

Welcome to Sun Yat-sen University night-time vehicle dataset! This dataset is aiming at night-time vehicle detection accuracy improvement. There are more than 5,000 images including over 12,000 night-time vehicles in two different traffic situations, i.e. urban road with road lamp and urban road without road lamp. All these images were collected by driving recorders placed on the front windshield. What’s more, vehicles numbers and each positions in each image is given in the annotation file. Some samples are shown as follows:

A benchmark toolkit is also included in the dataset, welcome to test your night-time vehicle detector on this dataset.


SYSU Dataset: Nighttime Vehicle (with Groundtruth)

Nighttime Vehicle Dataset contains more than 2000 images with rear of vehicle and real road scene at nighttime for vehicle detection.
The zip file includes the following:

  • 450 images with size 64*64 pixels as positive dataset, each image contains one rear of different kinds of vehicle with brilliant lamp in nighttime
  • 2000 images as negative dataset, containing nighttime road scene and other images in day time without vehicles
  • More than 1800 images for testing in training classifier, including positive and negative
  • 400 images with size 640*360 pixels for testing detection. These images taken in real road at nighttime contain 634 vehicles, and boundary box of all vehicles are recorded in groundtruth.txt file
  • README file
  • All images are in RGB and available in BMP format.

    The zip file is around 202MB, it can be downloaded in below link.