Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: afw
-
Labels:None
-
Story Points:6
-
Epic Link:
-
Sprint:AP F18-4, AP F18-5, AP F18-6
-
Team:Alert Production
Description
We've agreed that it should not be the responsibility of the PhotoCalib to include a Jacobian derived from the WCS, because we don't always want to apply that. Instead, SkyWcs should have the ability (given a pixel bounding box) to return a BoundedField that evaluates to the determinant of the Jacobian of the pixelToSky transform (aka the pixel area in angular units) at every point in that bounding box efficiently. This probably involves a new BoundedField class (though it's possible there's a way to do this with TransformBoundedField).
A reference implementation can be found in meas_mosaic's calculateJacobian functions, though a different approach to vectorization may be necessary to get good performance from AST.
Attachments
Issue Links
Activity
Reviews completed on the GitHub PRs. Minor comments in afw that should be addressed before merging. I don't understand how the changes in jointcal relate, since they don't use the new functionality in afw, but you say it's useful so I won't stand in your way; nevertheless, the commit messages should be fixed.
Paul Price
added a comment - Reviews completed on the GitHub PRs. Minor comments in afw that should be addressed before merging. I don't understand how the changes in jointcal relate, since they don't use the new functionality in afw, but you say it's useful so I won't stand in your way; nevertheless, the commit messages should be fixed. Thanks for the review.
Merged and done.
While reviewing fgcm, I discovered SingleFrameJacobianPlugin in meas_base plugins.py. It uses linearizePixelToSky().determinant (my version 2 above), so it should be very close to the pixel area calculation implemented here. However, it does result in the question of whether we need both of these approaches (the plugin itself doesn't appear to be used anywhere).
I suspect the plugin as it exists now would be better off using something based on getPixelScale() (and perhaps renaming it to make that clear). For completely different reasons (computing shapes and centroids in sky coordinates from pixel-coordinate measurements) we're going to also want to start saving all of the matrix elements returned from linearizePixelToSky independently, and I imagine we'll do that in the same plugin.
Paul Price: do you mind taking on this medium sized review (~350 new lines)?
The handful of new lines in jointcal were from my initial testing, but I figured they were useful enough to be added to the python layer.
If Jira doesn't pick up the afw PR, it's here: https://github.com/lsst/afw/pull/411