Details
-
Type:
Improvement
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: ts_aos
-
Labels:
-
Story Points:20
-
Epic Link:
-
Team:Telescope and Site
Description
Currently auxTel data is handled by default with generateDonutDirectDetect, which uses slow adaptive thresholding algorithm to detect donuts, not needing the WCS information. this was done because at the time auxTel WCS was insufficient to use directly to select sources based on the reference catalog. Quicker source detection codes (such as using the built-in https://github.com/lsst-ts/ts_wep/blob/develop/python/lsst/ts/wep/DonutDetector.py setting binaryChoice to centroid, or using the `afwDetect`) miss fainter blends with default settings, but could be used to obtain sufficient number of sources to solve for WCS. Given WCS, we can then update the donut catalog with https://github.com/lsst-ts/ts_wep/blob/develop/python/lsst/ts/wep/task/GenerateDonutCatalogWcsTask.py
This ticket is to investigate what components this task would require to work properly with auxTel data.
Attachments
Issue Links
- relates to
-
- Done
Activity
This was already tested, and we found that the best way forward was to implement a similar method as with `QuickFrameMeasurementTask`, which ended up being morphed into an updated `generateDonutDirectDetect`
Do you have a timing comparison between the new one and the old one just to show the difference?
I've run timing for the old one here https://jira.lsstcorp.org/browse/DM-37398 , I'll make a new notebook showing the timing of the new one to compare the two
Timing of the updated `generateDonutDirectDetect` task with simulated `lsstCam` and real `auxTel` data https://github.com/suberlak/AOS/blob/main/AOS_DM-37628_update_generateDonutDirectDetect.ipynb Previously it took ~2min 10 sec per image, now source detection only takes ~1.2 sec per image, i.e. 100x speedup.
New task added and new timing information shows speedup accomplished goals.
Is this complete with Bryce's work?