More colormap options!

With #90 @DanSchoppe added a third colormap to rio-tiler. While this is awesome, I think rio-tiler could do more for allowing user to define custom colormap.

colormap are used when translating a one band data array to the image format using the new array_to_image(using GDAL using a simple translation: https://github.com/cogeotiff/rio-tiler/blob/e512d8441e7ebb460d074a4a98a76051ad983b73/rio_tiler/utils.py#L386

So here ☝️ a rio_tiler.utils.array_to_image compatible colormap object is usually an array of shape (256, 3)

c = get_colormap(format="gdal")
print(c.shape)
 (256, 3)
print(c)
[
  [255 255 255]
  [250 250 250]
  [246 246 246]
  [242 242 242]
  ...
  [255   1 191]
  [255   1 207]
  [255   1 223]
  [255   1 239]
]

I first though we could have used a custom dict like

{
    0: [255, 255, 255],
    100: [255,   1, 191],
    ...
}

~to create a colormap fordiscrete values, but dict are not hashable so the current code in rio_tiler.utils.array_to_image won’t work.~

I do love the dict option for colormap (like Rasterio use https://github.com/mapbox/rasterio/blob/960a906dad2a4e426387ce048a52c6e90afdcd2b/docs/topics/color.rst#writing-colormaps) so I first check if we can maybe change the rio_tiler.utils.get_colormap function to have to return a dictionary colormap (because it seems to be the default format for GDAL) and in the same time adapt the rio_tiler.utils.array_to_image to work with dict.

In a second time we could add another utility function to create colormaps, it seems that matplotlib has some functionality but I’m not sure we should had more required module to rio-tiler.

[vincentsarago]

FYI: in #92 I’ve updated rio_tiler.utils.array_to_image to allow discrete colormap passed as dictionary

[vincentsarago]

[3 x 256] array is too large to pass as a query parameter, so if we stuck with our current colormap data structure, passing a full colormap at request time would probably be done via POST message body. At least it would compress well.

Nice thing about rio-tiler, is that it’s designed to be implementation agnostic to be used behind API or other functions (like rio-glui, remotepixel-tiler …). Passing a 3x256 array or dict via a POST message shouldn’t be a problem for you API. But I agree we need a tool which can create color map from a colormap definition (e.g. passing min/max color)

Perhaps there’s an elegant way to reduce the [3 x 256] array colormap schema to something more concise. You mentioned matplotlib, and I happened to find the same project for inspiration. If there’s a sufficiently concise way to describe a colormap, perhaps it would be palatable to include a colormap-producing description as a query param. Here’s the matplotlib colormap definitions for reference: https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/_cm.py I’m not sure there will be a flexible enough way to concisely describe colormaps without allowing the user to simply provide the full mapping for every index. For example, the colormap in #90 was produced by programmatically interpolating between colors in HSV colorspace.

Well as mentioned earlier, rio-tiler is agnostic, user could totally add Matplotlib to its API and apply the colormap using something else that rio-tiler.utils.array_to_image function. Meaning: I think at one point I’d like to keep rio-tiler as simple as possible so we focus on tiling performances and not on third party implementation (e.g: serving images).