Dataset - TrifoRobotics/IRONSIDES GitHub Wiki

We are pleased to release The Trifo Ironsides Dataset captured with our Ironsides device, a high-quality sensor and computing module designed for visual-inertial SLAM. The Ironsides is equipped with accurately synchronized and calibrated global-shutter stereo camera and IMU, and a multi-core processor. The dataset is captured in a robot arm platform as shown below.

This dataset is published in:

• Feng Zheng, Grace Tsai, Zhe Zhang, Shaoshan Liu, Chen-Chi Chu, Hongbing Hu, Trifo-VIO: Robust and Efficient Stereo Visual Inertial Odometry using Points and Lines, International Conference on Intelligent Robots (IROS 2018). [arxiv preprint] [Demo video]

Downloads

Dataset	Motion Type	Speed	Textureness	Difficulty	ROS bag	Zipped	Preview
PI_S1_X1	Pure X translation	Slow	Medium	Easy	Link	Link
PI_S1_X2	Pure X rotation	Slow	Medium	Easy	Link	Link
PI_S1_Y1	Pure Y translation	Slow	Medium	Easy	Link	Link
PI_S1_Y2	Pure Y rotation	Slow	Medium	Easy	Link	Link
PI_S1_Z1	Pure Z translation	Slow	Medium	Easy	Link	Link
PI_S1_Z2	Pure Z rotation	Slow	Medium	Easy	Link	Link
PI_S1_R1	Random	Medium	Medium	Medium	Link	Link
PI_3030	Random	Medium	Medium	Medium	Link	Link
PI_3058	Random	Fast	Low	Hard	Link	Link

Calibration and Groundtruth

• Calibration:
  • Download link: https://trifo-us.s3-us-west-1.amazonaws.com/Ironsides_Dataset_Release/calibration.tar.gz
  • Our calibration is defined with a tree structure using JSON file. IMU is the child of root. Left camera is the child of imu. And Right camera is the child of left camera. In each section, it includes id, model, calibration, transform etc.
• Groundtruth:
  • Format: timestamp tx ty tz qx qy qz qw
  • We recommend using TUM's benchmark scripts to do evaluation: https://vision.in.tum.de/data/datasets/rgbd-dataset/tools
  • Download link: https://trifo-us.s3-us-west-1.amazonaws.com/Ironsides_Dataset_Release/groundtruth.tar.gz