nvxSfmFindFundamentalMatNode fails to generate Matrix 'fundMat'

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

I am having two frame sources ‘L’ and ‘R’ (two different cameras i.e., one from each camera), After

vx_node nL = vxColorConvertNode(/parameters/);
vx_node nR = vxColorConvertNode(/parameters/);
vx_node nL1 = vxChannelExtractNode(/parameters/);
vx_node nR1 = vxChannelExtractNode(/parameters/);

vx_node nL2 = vxHarrisCornersNode(graph, imageLCY, strength_thresh, min_distance, sensitivity,
                                          harris_gradient_size, harris_block_size, Lharris_corners, Lnum_corners);
vx_node nR2 = vxHarrisCornersNode(graph, imageRCY, strength_thresh, min_distance, sensitivity,
                                          harris_gradient_size, harris_block_size, Rharris_corners, Rnum_corners);
vx_node nFM = nvxSfmFindFundamentalMatNode(graph,Rharris_corners,Lharris_corners,Foutput_keypoints,
                                          fundMat,errorThreshold,samples,indices,medianFlowThreshold);

When I tried to print fundMat by

vxReadMatrix(fundMat,checkerArray);
 std::cout << "fundMat: " << std::endl;
 i=0; while(i<9) {std::cout<< i << ": "<< checkerArray[i]; i++;}

It shows -1.07374e+08 in every index position for every frame. When I rendered harris_corners, both frames are doing fine. How do I fix this issue with nvxSfmFindFundamentalMatNode?

2

I’ve been playing with the SfM samples, too, and I don’t know exactly what the problem is in your case, but I have some questions:
Looking at the code in the SfM example, it looks like that function works on the “node” level and doesn’t actually do anything with the data directly.

Did you create the fundMat using vxCreateMatrix() first? (I assume so, but it doesn’t hurt to check)
What did nFM return when you created this node? (The sample uses NVXIO_CHECK_REFERENCE after allocation)
Did you run the graph before attempting to read data? (Again, I assume so, but it doesn’t hurt to check)

The SfM sample code never reads this matrix, it just uses it as a step towards the decomposed matrix node, from where it in turn reads translation and orientation.

If the creation of the fundamental matrix node fails, then perhaps you haven’t registered the SFM kernels before trying to create it? (Calling nvxSfmRegisterKernels() is needed I think.)
The example does it in main():
NVXIO_SAFE_CALL(nvxSfmRegisterKernels(context));

In general, I don’t see error checking in the code you post, so what do you learn if you print out the error/success of each call?