pixelated images about contextily

[lwasser]

You know, I've used ggmap a good bit but never looked under the hood to see how they are grabbing tiles. I'll dig a bit more to try to understand how it works so I can be more helpful here. Thanks so much. Leah A. wasser

[choldgraf]

+1 with sinc interpolation (you could always make this a kwarg)

[darribas]

This is cool. How could we integrate it in contextily? My sense is that this comes more at the plotting level than at the download, right? Maybe in plot_map? Incidentally, I've just realised that we have a calculate_zoom that @choldgraf implemented and I hadn't realised. It'd be great to get it as default for bounds2img and bounds2raster. But that's for a different ticket so will move that discussion over there :-)

[knaaptime]

sure, but i'm just the messenger. the credit goes to @ljwolf :)

so the quickest way to use it is to install from his fork until this gets merged

[ljwolf]

I won't speak for @darribas & am not a project maintainer, but I think that #21 may have stalled because of the auto-zoom request in #5, mentioned in #20.

Personally, I don't know if I'll have time to take a look at this until SciPy, but I'll try to un-stick #21 by then.

[darribas]

Quick update. @ljwolf and I are at SciPy and the idea is to cut a release that includes the interpolation, as well as some streamlining of the API (hopefully not back-breaking :) ). More to come soon!

[darribas]

This is now addressed in #32, so it's ready in master, should be getting out as a new release soon. If you need it urgently, you can pip install against master.

[choldgraf]

cc @lwasser as I know she was curious about this too

[lwasser]

ahhh yes @choldgraf @darribas i played around with the zoom and that didn't fix the issue which is when i create a larger figure, the map is then pixelated. Is this user error on my end?

[darribas]

Mmm... not directly, the way contextily works, every tile has the same size (248x248 if memory serves well) and the zoom controls how many tiles are used for a given extent. I'm not sure how you'd fix that or if it relates to contextily or the tile provider. Have you seen a different library working, for instance ggmap in R?

[jorisvandenbossche]

I was looking into this a bit, and I think doing interpolation in imshow is the best we can get to somewhat solve the pixelated view (it does not solve that you actually might want a higher resolution at a given zoom level ..).

Small example, for Berlin at zoom level 10, the original version (if you only do plt.imshow(basemap, extent=extent)):

And with interpolation='bilinear':

Which gives a bit better result I think (but as I said, does not give a higher resolution image, only makes the current one a bit nicer).
I think this is also what ggmap does, from a quick view on their source code (but no expert on that)

[choldgraf]

Nice - that would make sense to me as well...I was wondering how ggmap would get a smoother result from the same underlying data

[jorisvandenbossche]

@choldgraf see also https://github.com/darribas/contextily/issues/20, if you have an opinion on that

[ljwolf]

playing around with all the methods, I think sinc interpolation gives a marginally sharper image, over bilinear, since it appears to strongly preserve contrast better:


But, this is messing at the margins, since interpolation='none' looks pixelated regardless of setting axes=equal.

[lwasser]

Wow!! This is visually a lot nicer with the interpolation ! Happy to see this is being revisited. I still haven't looked closely at ggmap to understand how to works. Will this be implemented in contextily? If so I'll try it again for my course! Thank you all!

[ljwolf]

I think this would be what we set as defaults for #20

[lwasser]

hi all! i'm revisiting this again as i'm working on my class - again. just curious how or whether this might become the default behaviour for contextily plotting. For now i'm happy with using imshow to interpolate. Look at the difference!

BEFORE interpolation

after:

it would be a nice default plotting behavior if not already implemented. But i do appreciate and thank you all for demonstrating how the maps can be cleaned up using the imshow args.

If this has been implemented please let me know and we can also close this issue :)

[knaaptime]

the interpolation is included as default in 4850849 but hasn't been merged/released

[lwasser]

ok so install from github in the meantime as an alternative? Thank you @knaaptime ! This is a really nice fix.

[lwasser]

got it! ok thank you @ljwolf i can also just teach them to use the interpolation arg for the time being as well. i really appreciate everyone's work on this!

[lwasser]

oh wonderful. thank you for the update @darribas :) yay!

[lwasser]

i know this is closed but just writing to thank you again @darribas :)