Including ground truth camera pose during reconstruction

[eriksandstroem]

Dear everyone,
My aim is to use bundlefusion, but provide ground truth camera poses during reconstruction. I see that there is a variable which controls this called "s_binaryDumpSensorUseTrajectory" in the "zParametersDefaut.txt"-file. I can set that variable to true. I find that the variable is only called once. It is called in the DepthSensing.cpp file according to:

if (GlobalAppState::get().s_binaryDumpSensorUseTrajectory && GlobalAppState::get().s_sensorIdx == 3) {
      //overwrite transform and use given trajectory in this case
     transformation = g_depthSensingRGBDSensor->getRigidTransform();
     validTransform = true;

This requires using sensorIdx = 3, which corresponds to the so called BinaryDumpReader. It is not clear what the binary dump reader does compared to sensorIdx = 8 which reads .sens files offline. I can not make it run with sensor sensorIdx = 3.

If anyone knows how to deal with this, I would be very grateful.

Cheers,

[Uncle-Justice]

I got the similar issue. I wanna figure out how MatterPort3D dataset's mesh data is created by BundleFusion. I've never seen a image dataset like MatterPort3D as it is not made of continuous frames. But the authors did reconstruct the mesh very well. That is so magical. TAT