Since the method for determining areas of low confidence shows greater
promise than the method for oversampling those areas, two additional
nodes have been added, one to measure uncertainty in generation, and
another to extract a mask after generation has occurred.
Reducing the multiplier on resolution for oversampling seems to improve
output quality, but even in the best case, the effect of the node seems
little better than noise.
When interpolating the subrender location, the mode is now explicitly
set so it does not default to nearest.
Added code to terminate sampling on after first subrender to assist in
visualize the correctness of the operation.
Added code to allow more convenient setting of the weighting applied to
the subrender.
The founding concept, that the confidence of the output would be
measurable as a means of determining what areas of an image need further
processing is not bearing fruit. Saving state attempting more divergent
methods of measuring convergence.
Austin Mroz