Support writing PNGs with filter 3

[kkaefer]

Using filter 3 results in ~30% smaller PNG files in some situations.

[dnomadb]

@kkaefer It appears that GDAL http://www.gdal.org/frmt_various.html#PNG nor Pillow http://pillow.readthedocs.io/en/3.1.x/handbook/image-file-formats.html#png support that option; I added a webp output option that results in files ~1/2 the size, but at 2x+ the creation time.

[vincentsarago]

I added a webp output option that results in files ~1/2 the size, but at 2x+ the creation time.

I think it's because we use lossless=True, we could maybe use another config without impacting the quality too much.
e.g in mapnik we use default encoding is quality=75 and lossless=0. and in rio-tiler we use quality=80

[dnomadb]

because we use lossless=True

Yeah, which we have to, as any lossy compression or even resampling will corrupt the encoding.

[andrewharvey]

I wanted to chime in here from a users perspective higher up the Mapbox stack.

The first thing I noticed back when I tested out GL JS's hillshading was using raster-dem/terrain-rgb tiles were much slower to load than the original vector hillshading layers from the mapbox terrain tileset.

On a sample page the terrain-rgb tiles took up 3.2M of 4.1M in total, with the normal vector tiles poping in at about 1.5s, labels at about 2s, but then it takes until 6s for the terrain tiles to finish downloading. We should be able to do better than that.

So I took a look and found asking for terrain-rgb webp tiles only serves pngs.

https://a.tiles.mapbox.com/v4/mapbox.terrain-rgb/9/85/[email protected]
> identify [email protected]
PNG 512x512 512x512+0+0 8-bit sRGB

I tried a few compression options to see what affect that would have:

> for f in tiles/*; do b=`basename $f`; pngcrush -brute $f pngcrush/$b; done
> for f in tiles/*; do b=`basename $f`; cwebp -lossless -q 100 -m 6 $f -o cwebp/$b; done

method	total size
original	3.3M
pngcrush	2.4M ~30% improvement
cwebp	1.8M ~50% improvement

That matches exactly the findings of @kkaefer and @dnomadb given pngcrush I believe is finding the optimal png filter, both of them doing everything lossless.

Given webp isn't well supported that 30% gain on PNGs by changing the filter would be a huge win for users, together with the availability of webp terrain-rgb tiles for a 50% gain.

[andrewharvey]

Given webp isn't well supported that 30% gain on PNGs by changing the filter would be a huge win for users, together with the availability of webp terrain-rgb tiles for a 50% gain.

Bit off topic but I was happy to read this has been done now in the new Mapbox Raster Data API announced at https://www.mapbox.com/blog/mapbox-gl-js-v2-3d-maps-camera-api-sky-api-launch