In the AL implementation, the kernel candidates are small image stamps (footprints) around detected sources (on the already warped, pixel by pixel aligned images). The kernel candidate size depends on the kernel basis function pixel size (detecionConfig.fpGrowKernelScaling, fpGrowPix). Although the kernel candidates are always identified around sources (positions) on the science image, kernel basis function sizes were used inconsistently in case of science image convolution.

In the test runs, this did not make any substantial difference, though it could have caused problems with more extreme config parameter values. The kernel basis function pixel size differ in general if the images are swapped, i.e. switch between convolution and deconvolution cases in makeKernelBasisList.py: makeKernelBasisList().