EuRoC dataset evaluations

tum-vision / lsd_slam

LSD-SLAM

GNU General Public License v3.0

2.62k stars 1.23k forks source link

EuRoC dataset evaluations #329

Open jwangjie opened 5 years ago

jwangjie commented 5 years ago

Hello,

I was wondering has anyone successfully evaluated the LSD SLAM with the EuRoC dataset? I tested by rosrun lsd_slam_core dataset _files:=/EuRoC/V1_01_easy/mav0/cam0/data _hz:=20 _calib:=/euroc_cam0.cfg and rosbag play /EuRoC/V1_01_easy/V1_01_easy.bag using euroc_cam0.cfg as the calibration file. euroc_cam0.cfg follows the format specified in 3.1.3 Camera Calibration as:

458.654/752 457.296/480 367.215/752 248.375/480 0
752 480
none
752 480

The tracking lost soon after a few seconds for all EuRoC dataset (MH, V1, V2). Any suggestion?

Regards, Jie

pisuto commented 5 years ago

hi,

do you solve this problem? I have met the same issue.

jwangjie commented 5 years ago

hi,

do you solve this problem? I have met the same issue.

@pisuto Please try the following one, it works.

0.50355332 0.894045767 0.494061013 0.509863 0
1280 720
crop
400 288

I will post a comprehensive document talking about this later. Good luck.

pisuto commented 5 years ago

Hi, @jwangjie I use your file, but the ratio may be wrong. 400 288 this is warning: 2019-02-12 09 39 38

pisuto commented 5 years ago

I think this is what we need: 0.6099122 0.95061667 0.4883178 0.51722791 -0.28340811 0.07395907 0.00019359 1.76187114e-05 752 480 crop 640 480

glc12125 commented 4 years ago

Hey @pisuto , where did you get the 0.95061667 from? Following the README, we should put fy/height, which is 0.9527, as the second parameter. But 0.9527 gives worse result. Appreciate any insight.

I think this is what we need: 0.6099122 0.95061667 0.4883178 0.51722791 -0.28340811 0.07395907 0.00019359 1.76187114e-05 752 480 crop 640 480