MRL_Seq123_withSegmentation

Here is a set of three sequences recorded at the Mobile Robotic Laboratory (MRL) in ISR. The first bag is set to run up to 130 seconds, as a failure occurred after that point, but everything is correct until then. Along with sequence 1, there is a bag containing the semantic segmentation, obtained using the PSPNet network with the ADE20K dataset, with an opacity of 255. However, the filter.bag contains only part of the classes, while the all.bag document contains all the classes. Additionally, there are two more sequences with similar trajectories. The data includes recordings from the RGBD camera (D435i), wheel odometry, stereo-camera (Mynt Eye S1030), and 2D LiDAR (Hokuyo URG-04L). At this stage, the Mynt Eye did not yet have the IMU correction or the corrected right camera info values.

The transformations are defined as follows (approximately):

    RGB-D camera: 0.18 0.005 0.71 0 0.2007 0 base_link realsense_link
    Mynt Eye camera: 0.205 0.0 0.63 -1.57 0 -1.72 base_link mynteye_link

It should be noted that the laser transformations have already been recorded on the bag. The segmented images are in BGR format.
The ground truth (GT) was generated using RTAB-Map, combining odometry data from various sensors in a fusion module. While it is considered highly accurate, it may still contain some associated errors.