PRCV 2022: The FusionPortable-VSLAM Challenge
⏬ Dataset
| 🪧 Challenge
| 🏫 RAM-LAB
| 🧱 VisDrone
| 📧 Email
| 📝 Docs
| 📃 Paper
Sensor | Characteristics |
---|---|
3D LiDAR (not provided) | Ouster OS1-128, 128 channels, 120m range |
Frame Camera * 2 | FILR BFS-U3-31S4C, resolution: 1024 × 768 |
Event Camera * 2 | DAVIS346, resolution: 346 × 240,2 built-in imu |
IMU (body_imu) | STIM300 |
GPS | ZED-F9P RTK-GPS |
Ground Truth | Leica BLK360 Imaging Laser Scanner |
The submission will be ranked based on the completeness and frequency of the trajectory as well as on the position accuracy (ATE). The score is based on the ATE of individual points on the trajectory. Points with the error smaller than a distance threshold are added to your final score. This evaluation scheme is inspired by HILTI Challenge.
Output trajectories should be transformed into the body_imu frame, We will align the trajectory with the dense ground truth points using a rigid transformation. Then the Absolute Trajectory Error (ATE) of a set of discrete point is computed. At each ground truth point, extra penalty points are added to the final score depending on the amount of error at this point:
Error | Score (points) |
---|---|
<= 5cm | 10 |
<= 30cm | 6 |
<= 50cm | 3 |
<= 100cm | 1 |
> 100cm | 0 |
Each sequence will be evaluated over a maximum of 200 points, which leads to a maximum of $N\times 200$ points being evaluated among $N$ sequences.
Given an example:
Sign up for an account and submit your results in the evaluation system, the live leaderboard will update your ranking.
Trajectory Results
traj/20220215_canteen_night.txt
traj/20220215_garden_night.txt
traj/20220219_MCR_slow_00.txt
traj/20220226_campus_road_day.txt
....
1644928761.036623716 0.0 0.0 0.0 0.0 0.0 0.0 1.0
....
Each row contains timestamp_s tx ty tz qx qy qz qw. The timestamps are in the unit of second which are used to establish temporal correspondences with the groundtruth. The first pose should be no later than the starting time specified above, and only poses after the starting time will be used for evaluation.
T_bodyw_body = T_body_sensor * T_sensorw_sensor * T_body_sensor^(-1);
.Do not publicly release your trajectory estimates, as we might re-use some of the datasets for future competitions.
A team can only register one account. Quota can only be obtained by joining the WeChat group. In order to prevent the problem of a team registering multiple accounts, this competition requires all members of the participating team to join the WeChat group. And the old account cannot be used, you need to re-register a new account.
All data download addresses can be found in this directory :📁
We provide the compressed rosbag data, remember to execute the following command to decompress them.
# example: 20220216_garden_day_ref_compressed
rosbag decompress 20220216_garden_day.bag
Yaml Files | Describtion | Link |
---|---|---|
body_imu | extrinsics and intrinsics of the STIM300 | body_imu.yaml |
event_cam00 | extrinsics and intrinsics of the left event camera | event_cam00.yaml |
event_cam00_imu | extrinsics and intrinsics of the left event camera imu | event_cam00_imu.yaml |
event_cam01 | extrinsics and intrinsics of the right event camera | event_cam01.yaml |
event_cam01_imu | extrinsics and intrinsics of the right event camera imu | event_cam01_imu.yaml |
frame_cam00 | extrinsics and intrinsics of the left flir camera | frame_cam00.yaml |
frame_cam01 | extrinsics and intrinsics of the right flir camera | frame_cam01.yaml |
Platform | Sequence | Compressed Bag | Ground Truth | |
---|---|---|---|---|
Handheld | 20220216_garden_day | 20.4GB | 20220216_garden_day.txt |
Platform | Sequence | Compressed Bag | |
---|---|---|---|
Handheld | 20220209_StaticTarget_SmallCheckerBoard_9X12_30mm | 6.7GB | |
Handheld | 20220215_DynamicTarget_BigCheckerBoard_7X10_68mm | 2.3GB | |
Handheld | 20220209_Static_IMUs_3h20mins | 894MB |
Platform | Sequence | Compressed Bag | |
---|---|---|---|
Handheld | 20220216_canteen_night | 15.9GB | |
20220216_canteen_day | 17.0GB | ||
20220215_garden_night | 8.5GB | ||
20220216_garden_day | 20.4GB | ||
20220216_corridor_day | 27.4GB | ||
20220216_escalator_day | 31.7GB | ||
20220225_building_day | 37.5GB | ||
20220216_MCR_slow | 3.5GB | ||
20220216_MCR_normal | 2.2GB | ||
20220216_MCR_fast | 1.7GB | ||
Quadruped Robot | 20220219_MCR_slow_00 | 9.7GB | |
20220219_MCR_slow_01 | 8.4GB | ||
20220219_MCR_normal_00 | 7.1GB | ||
20220219_MCR_normal_01 | 6.5GB | ||
20220219_MCR_fast_00 | 7.6GB | ||
20220219_MCR_fast_01 | 8.5GB | ||
Apollo Vehicle | 20220226_campus_road | 72.3GB |
The picture below is a schematic illustration of the reference frames (red = x, green = y, blue = z):
We will provide some sample datasets along with their ground truth collected with the same sensor kit, but the ground truth for the challenge sequences is not available. However, you can submit your own results in the website evaluation system for evaluation.The ground truth for all challenge sequences will finally be announced at the PRCV WORKSHOP in October.
When using this work in an academic context, please cite the following publication:
@article{,
author = {Jianhao Jiao and Hexiang Wei and Tianshuai Hu and Xiangcheng Hu and Yilong Zhu and Zhijian He and Jin Wu and Jingwen Yu and Xupeng Xie and Huaiyang Huang and Ruoyu Geng and Lujia Wang and Ming Liu},
title = {FusionPortable: A Multi-Sensor Campus-Scene Dataset for Evaluation of Localization and Mapping Accuracy on Diverse Platforms},
booktitle = {2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2022}
}
This challenge was supported by the Wireless Technology.
We would like to thank the AISKYEYE Team at Lab of Machine Learning and Data Mining of Tianjin University, for hosting our challenge at the PRCV2022 workshop. Futher, this challenge would not have been possible without the assistance of Prof.Ming Liu, Prof.Lujia Wang, Prof.Pengfei Zhu, Prof.Dingwen Zhang, Dr.Zhijian He and Dr.Jianhao Jiao for the great support in organizing the challenge, verifying the data and providing the HILTI Challenge 2022 as template for this challenge.
We would also like to thank Prof.Jack Chin Pang CHENG and his team for the support of dense mapping device.
All datasets and benchmarks on this page are copyright by us and published under the Creative Commons license (CC BY-NC-SA 3.0), which is free for non-commercial use (including research).