Some Questions

[maffan-96]

Hello,
Thanks for such a wonderful work. I am wondering if you could direct me to the mesh-merging implementation/code for better understanding as it is not very clear from the paper.
Regards

[Jianhao-zheng]

Hi, thank you for your interest in the work. Sure, the core mesh-generating code is in this function.

MAP-ADAPT/voxblox/include/voxblox/mesh/mesh_integrator.h

Line 416 in 03c9d9a

void extractBlockMeshAdaptive(typename Block<VoxelType>::ConstPtr block,

[maffan-96]

Thanks for your quick response. Can you also refer me to the TSDF/voxel subdivision place (voxel map representation)? (Referring to the paper, it states that voxels are sub-divided when a certain threshold is met) where exactly the decision has been taken.
Secondly, the paper also described that ray casting is modified to suit adaptive resolution such that three virtual grids are created for respective resolution. However, I am confused as to the best of my understanding there is only one TSDF map. So, does the voxels formed from those three virtual grids are integrated to form one unified multi-resolution TSDF map?
Thank you

[Jianhao-zheng]

Hi, the decision is taken here:

MAP-ADAPT/voxblox/src/integrator/tsdf_integrator.cc

Line 761 in 03c9d9a

if (adaptive_mapping) {

.
Other function also uses similar logistics, but it's the most representative one.

For your second question, the virtual grid is not a reconstruction map. It's "virtual" as we only use this to decide whether we should skip the ray to save running time. See this line https://github.com/GradientShttps://github.com/GradientSpaces/MAP-ADAPT/blob/03c9d9a710493b01ef510d0df2a26b025e668efb/voxblox/src/integrator/tsdf_integrator.cc#L1773paces/MAP-ADAPT/blob/03c9d9a710493b01ef510d0df2a26b025e668efb/voxblox/src/integrator/tsdf_integrator.cc#L1773
We will only inject the information only if the virtual grid is not occupied by any previous ray. But we do not do any geometric/color reconstruction on virtual grid.