All `camcalib` and `imucalib` datasets were recorded with a fixed exposure of
3000μs and a gain of 32/255 on the OV7251 sensor looking at aprilgrids with
square side of 3cm. For the magnetometer `magcalib` datasets the exposure/gain
value adjusts automatically per camera like in the MOO datasets. All datasets
except MOC13 were recording on the same day, and on the same Monado euroc
recording session.

1. MOC01-MOC06: Were recorded at the beginning of the session
2. MOC06-MOC12: Were recorded at the end of the session. The mocap calibration
   of these datasets have a big difference in "pz" compared to the previous
   MOC01-MOC06 ones. I believe this is due to steamvr readjusting the floor
   height at some point during runtime since I don't have reason to believe I
   moved the lighthouses. Furthermore using any of the two mocap versions when
   doing mocap-imu alignment produces datasets with similar ATE after trajectory
   alignment.
3. MOC13: Is a 48hs recording of the headset still. It was recorded at a
   different date and only has raw IMU samples for Allan variance analysis. The
   groundtruth data is manually crafted with just two points in the same
   position.

The Reverb G2 headset has 4 cameras:
- 0: front-left
- 1: front-right
- 2: side-left
- 3: side-right

[Basalt calibration
tools](https://gitlab.com/VladyslavUsenko/basalt/-/blob/master/doc/Calibration.md)
were used for generating the calibration files.

Magnetometer timestamps should be hardware alignmed with the other IMU samples
(imu0). Consider that, at the time of recording, the magnetometer units the
headset reports was not yet definitely known in the open-source WMR Monado
driver. What you will find for the units is a guess, and it might change in the
future if the Monado community figures out the units were wrong. For
magnetometer calibration you can look up this video guide:
https://www.youtube.com/watch?v=MinV5V1ioWg