[QGIS-Developer] QGIS Georeferencer improvements

[Nyall Dawson]

On Wed, 1 Apr 2020 at 07:27, Martin Dobias <wonder.sk at gmail.com> wrote:

> I have been thinking about this and what are the pros and cons of
> having a separate window for georeferencing vs doing georeferencing in
> main map canvas. Actually rather than discussing whether to have one
> or two windows, the key bit to decide on is something else - what
> would be the coordinate system of the georeferencer canvas:
>
> A. image coordinates - that is what we have in QGIS georeferencer now:
> the image gets loaded without any distortion and stays that way all
> the time. Picking of map coordinates for GCPs has to be done in a
> separate map canvas. (i.e. one click in "georeferencer canvas", and
> one click in "reference map canvas")

Well - kind of. We do some stuff in image coordinates, yet confusingly
require a CRS selection when first loading a raster (which I believe
is used for nothing at all), and then the map coordinates of GCPs have
an innate CRS too (which can be changed when you actually save the
image, causing all sorts of weirdness!)

> B. map coordinates - that is what ArcGIS does. The image is placed at
> some initial position in the map, picking of image coordinates and map
> coordinates for GCPs is done in the same map canvas. (i.e. two clicks
> in single map canvas)
>
> The option B has the advantage that it is quite easy/intuitive to have
> live preview of the image transformation as GCPs are being added. With
> option A the live preview of the transform is not really possible (we
> are still in image coordinates) - and the preview would need to be in
> a separate map canvas.
>
> So where I am hesitating is whether to stick to option A which is
> probably familiar to QGIS users - or with option B that may provide
> better UX thanks to the live update of the warped raster (or vector).
> I guess we will need to choose one or the other - supporting both
> options would be probably too confusing. I am trying to figure out
> which way to take, because this will also have great impact on the
> workflow on how GCPs would be picked on the map(s) using map tools...

I think option B makes much more sense. It may act different to what
users are used to, but my experience is that users find the current
approach confusing. I certainly do myself, and that's a bad sign. I've
also wasted a lot of time over the years fixing mine/others
georefencing errors caused by the confusing mix of CRS options in the
current georeferencer.

Reasons I like option B:
- the live updates you've mentioned above
- we know exactly what CRS already recorded GCPs are in, so we can
easily transform them to a different CRS if at time of saving the user
does decide to pick a different CRS vs what they've done the
registration in

My one question: where do you initially place the image? Do you just
throw it in the middle of the current canvas extent as a starting
point?

Nyall



>
> Thanks for your thoughts :-)
>
> Cheers
> Martin