Fix shifts - Githubissues

herjy commented 4 years ago

This PR fixes a few bugs that were previously affecting the alignment of multiresolution frames.

In particular it preforms a needed axis inversion by inverting the sign of the argument of a shift, which is the correct mathematical way of doing it here, instead of inverting the axis of an array.
The other source for unwelcome shifts was the inconsistency between even and odd shape paddings of images at different resolutions.
Another correction that seems to have no effect was to avoid setting arguments of a complex exponential as an axponent. This should generally be avoided as complex exponentials do not behave in the same way as the exponential function in R.

pmelchior commented 4 years ago

Can you check the tests, please

herjy commented 4 years ago

shifts were handled inconsistently between the adam and the fix_shifts branch. Fixed now.

pmelchior commented 4 years ago

Please remove the file docs/hsc_cosmos_35.sca. It should not be tracked.

pmelchior commented 4 years ago

I'll do a few more checks later today, so expect a few comments then.

pmelchior commented 4 years ago

LowResObservation.match seems to have slowed down considerably. I remember taking only 200ms, and it's now around 1s. Do you know the reason? Or am I misremembering?

pmelchior / scarlet

Fix shifts #155