Passing a vector object to Matlab trough Mex routine

I have an definition of a vector as typedef std::vector<Ipoint> IpVec;

Ipvec ipts; //this is the object i Use.

Ipoint is a class which has the following variables:

float x, y;
float scale;
float orientation;
int laplacian;
float descriptor[64];
float dx, dy;
int clusterIndex;

How do I return ipts back to Matlab ?

Answer 1

[Geoff]

You need to build the struct yourself. This example might help point you in the right direction:

http://www.mathworks.com/help/techdoc/matlab_external/f12977.html#f11333

If you don't have the code, I found an old version here:

http://sipi.usc.edu/manuals/matlab701/techdoc/matlab_external/ch04cre8.html

While I haven't tried this, and I'm not sitting in front of MatLab or a C++ compiler right now, I don't expect it would be a problem.

Try something like this to start:

const int nfields = 9;
const char * fieldnames[] = {
  "x", "y",
  "scale",
  "orientation",
  "laplacian",
  "descriptor",
  "dx", "dy",
  "clusterIndex"
};

plhs[0] = mxCreateStructMatrix( ipts.size(), 1, nfields, fieldnames );

for( int i = 0; i < ipts.size(); i++ ) {
  lpoint & p = ipts[i];
  mxSetFieldByNumber(plhs[0], i, 0, mxCreateDoubleScalar(p.x));
  mxSetFieldByNumber(plhs[0], i, 1, mxCreateDoubleScalar(p.y));
  mxSetFieldByNumber(plhs[0], i, 2, mxCreateDoubleScalar(p.scale));
  // etc...
}

You may want to make a helper functions for creating and copying the 64-elem descriptor array, as well as something to explicitly create integer scalars if you don't want them to be converted to doubles.

[Geoff]

Oh yes, if that doesn't work you can take the approach from the phonebook example, which creates a 1x1 struct matrix and then creates an array (or a matrix in the case of the descriptor field) for each field. That would probably be more efficient, actually.

Answer 2

[Rohit]

I am sorry I wasnt able to follow you. the following error comes : error C2440: 'initializing' : cannot convert from 'Ipoints' to 'ipts &' error C2039: 'x' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(890) : error C2039: 'y' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(891) : error C2039: 'scale' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(892) : error C2039: 'orientation' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(893) : error C2039: 'laplacian' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(894) : error C2039: 'descriptor' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(895) : error C2039: 'dx' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(896) : error C2039: 'dy' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ] main1.cpp(897) : error C2039: 'clusterIndex' : is not a member of 'std::vector<_Ty>' with [ _Ty=Ipoint ]

Also I wasnt able to understand creating helper functions. i am pretty new to mex.

[Geoff]

Oh, yes my code should say 'lpoints', not 'lpVec'. I'll edit the original. I assume that while you are new to MEX, you are familiar with C++ and the STL? Do bear in mind I wrote this snippet without sitting in front of MatLab or a compiler, and I haven't used mex's structs before. So take my answer as a reasonable starting point, but also with a grain of salt! =)

[Rohit]

I tried it, but the following error comes.

error C2664: 'mxCreateDoubleScalar' : cannot convert parameter 1 from 'float [64]' to 'double'

There is no context in which this conversion is possible

If i remove that line then these errors come

main1.obj : error LNK2019: unresolved external symbol _mxSetFieldByNumber_700 referenced in function _mexFunction

main1.obj : error LNK2019: unresolved external symbol _mxCreateDoubleScalar referenced in function _mexFunction

main1.obj : error LNK2019: unresolved external symbol _mxCreateStructMatrix_700 referenced in function _mexFunction

main1.mexw32 : fatal error LNK1120: 3 unresolved externals

[James Tursa]

Can you post the code you currently have for us to look at? The error C2664 above sounds like you passed a float array into the routine instead of a float