Closed djkapner closed 6 years ago
rebased to master
Merging #176 into develop will not change coverage. The diff coverage is
n/a
.
@@ Coverage Diff @@
## develop #176 +/- ##
========================================
Coverage 96.47% 96.47%
========================================
Files 57 57
Lines 3550 3550
========================================
Hits 3425 3425
Misses 125 125
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update c37afe4...13f43b8. Read the comment docs.
…rization
EMaligner now has finer tuned control for polynomial parameter regularization. For example, for 2nd order polynomial, a regularization.poly_factors list of
[1e-5, 1.0, 1e6]
specifies factors for 0th, 1st, and 2nd order parameters respectively. Each of these are multiplied by regularization.default.lambda, keeping with tradition of having default_lambda and translation_factor for affine, which is equivalent to a 1st order polynomial with poly_factors[translation_factor, default_lambda]
The purpose is to add some higher constraint to the 2nd order parameters, as, for poorly matched point matches, these parameters can get large enough to drastically distort a tile, usually resin or blood vessel. Specifying something like 1e6 prevents these higher order corrections from becoming too powerful.