I can confidently say that sufficient overlap was not the issues. I experimented a lot, but I definitely had well over 99% overlap on some runs without really

I can confidently say that sufficient overlap was not the issues. I experimented a lot, but I definitely had well over 99% overlap on some runs without really any improvement. I suspect it has to do with the training data used for feature matching.