woodpeck
commented
12 years ago I was about to open a new issue when I found this old one which has perhaps been sidelined because it is not very specific. I am fighting raster reprojection on a regular basis and here's a detailed example, based on SRTM hillshading.
I have the following style that simply displays hillshading and nothing else:
<Map background-color="#fff" srs="+proj=merc +a=6378137 +b=6378137 +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0 +k=1.0 +units=m +nadgrids=@null +wktext +over +no_defs">
<Style name="hillshade">
<Rule>
<RasterSymbolizer opacity="0.4" scaling="bilinear8" mode="normal" />
</Rule>
</Style>
<!-- SRS identical to above -->
<Layer name="srtm" status="on" srs="+proj=merc +a=6378137 +b=6378137 +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0 +k=1.0 +units=m +nadgrids=@null +wktext +over +no_defs">
<StyleName>hillshade</StyleName>
<Datasource>
<Parameter name="type">gdal</Parameter>
<Parameter name="file">hillshade.tiff</Parameter>
<Parameter name="format">tiff</Parameter>
</Datasource>
</Layer>
</Map>The source image for this test is a 32x32 image cropped from SRTM hillshading, shown magnified to 256x256 here:
The desired result, and indeed the result with the current latest 2.1.0+dev20120816 build, is this:
However, when I add a blank to the SRS definition in the <Layer> tag and thereby force Mapnik to do a raster reprojection (even if the mathematical outcome of that reprojection is a null operation), I get this result with ugly white speckles:
(NB, this is an improvement over 2.1.0+dev20120703 which skipped smoothing altogether in this situation and created a very blocky output similar to the raw TIFF posted at the top.)
My issue with this is not that Mapnik runs the projection code at all, not seeing that both projections are indeed the same; I chose two different-looking-but-identical projections for testing only. My issue is that I actually want to use Mapnik in a situation where raster reprojection is not avoidable (namely, a WMS server that suports dozens of projections and it would be a major hassle to pre-project every raster into every possible projection).
I have prepared a little manual test case consisting of a shell script that you can run to produce the above "correct" and "broken" results with nik2img, here: http://www.remote.org/frederik/tmp/hillshadeprojection.zip - would be interesting to hear if others can confirm the problem or maybe even have a workaround (that is different from "just use gdalwarp...").
springmeyer
Andrey-VI
dieterdreist
Resampling doesn't work correctly for gdal images which are being reprojected on the fly. The work around for now is to use gdalwarp to adjust them to your map's projection before rendering.