Retrieving the uncertainty of a vertex in g2o

edit

I am manipulating the tutorial_slam2d example in the g2o package. I would like to output the marginal uncertainty of each vertex. However, using the following code snippet produces results that don't look right:

SparseOptimizer optimizer;
...
ofstream g2oOptResults;  
string g2oFileID = "g2oResults__" + fileID + ".txt";

g2oOptResults.open(g2oFileID.c_str());

for (int i=0; i<5; i++)
{  
VertexSE2* testPose = dynamic_cast<VertexSE2*>(optimizer.vertex(i)); 
Matrix<double,3,3> uncertainty = testPose->uncertainty();

g2oOptResults << uncertainty << endl;  
g2oOptResults << " " << endl;
}

g2oOptResults.close();

The problem is the returned matrix clearly cannot represent a covariance matrix; it is not symmetric. As a typical example I get:

0.325765, 0.919874, 0.919874
-0.00215401, 0.00202767, 0.00202767
0.0196631, 0.00424371, 0.00424371

The values look like they might belong to a symmetric matrix (i.e. I get three unique values and three pairs of identical values). They are just not arranged correctly.

Any thoughts?

Edit: on a related note, the vertices in optimizer (instance of SparseOptimizer) have an uncertainty associated with them _before_ optimization. It doesn't seem that should happen; I don't tell optimizer anything about uncertainty in the vertices, just the fixed edges.

(Sorry this is so lengthy)