Generating DS for "dense_slam_gui.py"

[brleme]

Hi, everyone!

I am trying to generate a dataset for using with the "dense_slam_gui" example.
I use a realsense camera D455 and generate the files in two folders.

When I run the slam, the visual odometry gets very unstable and start messing the whole point cloud.
Until now I checked:

• Tried different resolutions for the color and depth images
• Checked the channels and type of my data (color 3 channels 8bits, depth 1 channel 16bits)
• Changed realsense parameters (randomly)

It is strange that even when the camera is stopped (or with a very small movement) my point cloud starts to shift.
I am trying to change parameters for recording the images in the Realsense camera but no success so far.

To avoid complexity, I am running the example code as it is. Also, I tried different datasets (from the internet) and they are working amazing (what reinforce that my ds should be the problem)

This is an example of one of my best reconstructions.

• please let me know whether I should upload more material.

My ds:

I would really really appreciate any help.

Thank you,