Investigate further inlier ratio w.r.t. GT model and w.r.t estimated model

[johnwlambert]

Currently, the inlier ratio with respect to GT model is greater than that with respect to estimated model for USAC. This is perhaps expected, as the GT should explain the data most appropriately. But our inability to estimate a model with close to as many inliers is unfortunate, and perhaps could be explored further on some representative image pairs.

We see this with USAC on Skydio-8:

However, we do not see this with LORANSAC on Skydio-8

cc @akshay-krishnan @ayushbaid

[akshay-krishnan]

It is expected to have a high inlier ratio for the GT model (if the matcher output is good), but I don't understand why the inlier ratio with respect to the estimated model is lesser than the GT model.
The GT model was not estimated from our data, and if we had enough verifier iterations, we should have the model with the highest number of inliers. I'm not familiar with how USAC works though.

Another reason this could happen is if when we have degenerate cases (all inliers on a plane) and our estimated model is not good, but skydio-8 is not such a dataset right?