Skersky Dataset: The image mosiacking after GTSAM optimization still has room for improvement

[KSorte]

Description

After the commit #9 , the major issues around the image mosiacking after GTSAM optimization such as overly conservative covariances and thresholds were mitigated.
However, the stitching is still not perfect and as such may perform poorly for the 180 image dataset.

Considerations

1 For example, in the 29 image set, the first and second column seem not to get enough matches in non temporal link formulation. Look at image 6. Getting better links there should better the stitching.
2. Explore better strategies for non temporal link proposal and validation and better covariances.
3. Inspect if the overlaps between the non temporal links are good.
Explore if multiple rounds of optimization are needed.
4. Can Non Maximal Suppression be implemented anywhere here for better feature detection?
5. However, unless meticulous tuning is undertaken, perfect mosiacking may not be achieved because optimization is carried out only with respect to the image centers and the not features which the Pizzaro 2003 paper does.