Open wmwv opened 3 months ago
Images have already been ISRed and have catalogs, WCS, PSF, and photometric calibration (which we might use for normalization).
INPUT: Image 1, Variance 1, Catalog 1, PSF 1 Image 2, Variance 2, Catalog 2, PSF 2
I think the steps are
OUTPUT: Score image, Subtracted image, Convolution kernel, [Co]Variance image + matrix.
Detection happens after.
Rewrite the individual steps to pass CuPy arrays instead of writing out and reading from disk for each intermediate step. Along the way we'll write down clearly what each step needs as input and output and what metadata we'll want to pass along to the final output even if it's not used by any intermediate step (e.g., MJD-OBS, FILTER, EXPTIME among other things).
A longer-term goal is to consider combining matrix operations from multiple steps and doing. But this issues is just about passing things in memory following the current steps.