./Register gives segmentation error

[iszihan]

Hello, I am trying to align two spherical meshes that I got from running ./SphereMap but it gives me segmentation errors. After looking into the codes, it seems the segmentation error happens here

MoebiusRegistration/Register/Register.cpp

Line 209 in c726787

for( int i=0 ; i<targetTessellator.polygons().size() ; i++ ) targetCells[ p[i].theta*Resolution.value + p[i].phi ].push_back( p[i] );

in which p.theta and p.phi gives value of 2147483647, which seems to be out of range. I didn't have much success tracing this behavior, and therefore want to submit an issue to see if there is a bug that can be fixed.

Please see https://drive.google.com/drive/folders/1Q_8BltOTxNC6yIg6jf4AaXcwgvXvAloX?usp=share_link for the two spherical meshes I was trying to align if this is helpful. Thank you!

[mkazhdan]

Are you sure these are the output of SphereMap?
These do look like the results of spherical parameterization, but they only have the spherical positions, not the original vertex positions. (Typically the output of SphereMap should have the original vertex positions in the "x", "y", and "z" fields and the spherical positions in the "px", "py", and "pz" fields.)

It could be that you accidently ran the "Parametrization" executable which replaces the original vertex positions with the spherical positions...

[iszihan]

Thank you for the response. I also uploaded the original meshes which I ran with ./SphereMap --in 100.ply --out 100_cmcf --spherical , and I made sure I wasn't running ./Parameterize. Was there something wrong with these specified options?

[iszihan]

I tried running without the --spherical option and when I tried to open the output file with libigl, it says
ReadPLY error: all requested properties must share the same type

[mkazhdan]

You shouldn't be using the "--spherical" option as that replaces the original vertex positions with the spherical positions (you need both for registration).

[mkazhdan]

I'm afraid that's an issue with libigl. (It's not happy that "x", "y", and "z" are single precision while "px", "py", and "pz" are double precision.)
Unfortunately, you do want to be using double precision for the sphere map as there can be significant scaling the cMCF flow.

If you try opening the file with MeshLab, things should appear OK.

[iszihan]

Thank you and there is no segmentation error anymore! However, when I run ./Register with --out option it doesn't give me an output file -- is that expected?

[mkazhdan]

That's unexpected. Can you share the original files?

[iszihan]

Yes I uploaded the new outputs from ./SphereMap which I passed in as input to ./Register to https://drive.google.com/drive/folders/1Q_8BltOTxNC6yIg6jf4AaXcwgvXvAloX?usp=share_link , which are 100/300_cmcf_rightone.ply : )

[mkazhdan]

Strange I ran:
SphereMap.exe --in 100.ply --out A.ply
SphereMap.exe --in 300.ply --out B.ply
and then:
Register.exe --in A.ply B.ply --out foo.ply
and got a file foo.ply with header info:
ply
format binary_little_endian 1.0
element vertex 6890
property float x
property float y
property float z
property double px
property double py
property double pz
property uchar red
property uchar green
property uchar blue
element face 13776
property list int int vertex_indices
end_header

[iszihan]

I was running this on Linux with a different Makefile from #2 -- would that be the problem?

[mkazhdan]

Running with the latest Makefile on WSL (ubuntu) I get:

SphereMap --in 100.ply --out A.ply --verbose
CMCF[100] 12.15(s): D-Norm=0.00030 / QC-Ratio=1.21715 / R-Deviation=0.00190
M-center[3] 0.01(s): 6.56e-11 (-1.8e-12 -6.6e-11 -1.0e-12)
SphereMap --in 300.ply --out B.ply --verbose
CMCF[100] 14.94(s): D-Norm=0.00043 / QC-Ratio=1.21550 / R-Deviation=0.00199
M-center[3] 0.01(s): 5.37e-11 (-6.2e-12 -5.3e-11 -3.8e-12)
Register --in A.ply B.ply --out AB.ply --verbose
Tessellated: 0.65(s)
Correlated 1.53(s) Error: 0.0296604

What happens when you add the "--verbose" flag? Do you still the "Correlated ... Error ..." line?