I have been testing ORB-SLAM2 with some Visual Odometry datasets.

In some datasets ORB-SLAM2 losses track when the ROS bag is played in normal speed. However, if I play the ROS bag in 20% of its original speed the algorithm finishes the trajectory estimation without any issue.

I have measured the time that my laptop needs to process every frame with ORB-SLAM2 and it is far more capable than required (ROS bag frames at 10 fps, while my laptop can handle them at 34 fps). I really can't figure out what the issue is, and by the time my processor does not choke, I can't understand why it looses track if I play the ROS bag in normal speed.