Open Skoltech Car
Dataset files
Dataset sequences can be downloaded from http://oscar.skoltech.ru/data/.
Note that currently only a small portion of dataset is uploaded.
Each sequence is stored in a separate folder, which can contain the following files:
left.tar
andright.tar
: frames from left and right cameras respectively compressed using lossless algorithm. Each file inside tar archive is named as<unix timestamp, us>_<monotonic clock, us>.flif
. For converting frames to PNG, PNM or JPEG, you can useconvert
utility from here. Please referer to the [paper][icra2019_paper] for more details.left_jpgs.tar
andright_jpgs.tar
: frames with applied histogram equalization filter compressd using JPEG lossy compression.left_video.webm
andright_video.webm
: video assembled from frames after with applied histogram equalization.velodyne.pcap.xz
: compressed network packets from Velodyne HDL-64E S3 LiDAR.imu.bin.xz
: InertiaCube3 IMU data.gps.log.xz
: GPS NMEA messages from Garmin 18x receiver.
Annotations
Traffic sign annotations can be downloaded from https://github.com/icevision/annotations/.
Citations
If you use this dataset in your research, please cite the following paper:
![]() |
Artem L. Pavlov, Pavel A. Karpyshev, George V. Ovchinnikov, Ivan V. Oseledets, and Dzmitry Tsetserukou, "IceVisionSet: lossless video dataset collected on Russian winter roads with traffic sign annotations", IEEE International Conference on Robotics and Automation (ICRA), 2019. [PDF] |