Shifted tiles with mbtiles

[ottoxt]

After looking into the source code of gdal (http://trac.osgeo.org/gdal/browser/trunk/gdal/frmts/mbtiles/mbtilesdataset.cpp) I've decided my question of stackexchange suites better here:

I'm confused because of the representation of mbtiles files in QGIS. It seems that the whole grid is shifted by about 3 m to the south. So far I can observe this behavior only in the context of QGIS and mbtiles. What I did: Loaded Geotiff in Tilemill and exported as mbtiles. Loaded both files in QGIS. See result in attachment.

On the search for the cause I have

• generated the mbtiles file with geoscript-groovy: same result -> Tilemill does not seem to be the cause.

• taken an other GeoTIF -> same result: shifted 3m south.

• loaded the mbtiles-file in MapProxy, published it as WMS and loaded it in QGIS -> fits perfectly: the mbtiles file itself it seems so not to be the cause.

[ottoxt]

Taken from QGIS: Original TIF on top. Bottom with transparent mbtiles overlay.

[Even Rouault]

Could you try the following patch ?

$ svn diff frmts/mbtiles/
Index: frmts/mbtiles/mbtilesdataset.cpp
===================================================================
--- frmts/mbtiles/mbtilesdataset.cpp	(révision 28173)
+++ frmts/mbtiles/mbtilesdataset.cpp	(copie de travail)
@@ -1049,8 +1049,7 @@
 /*                          GetGeoTransform()                           */
 /************************************************************************/
 
-//#define MAX_GM 20037508.3427892
-#define MAX_GM 20037500.
+#define MAX_GM 20037508.3427892
 
 CPLErr MBTilesDataset::GetGeoTransform(double* padfGeoTransform)
 {

Are you sure the shift is 3m and not 8m ?

[ottoxt]

Replying to rouault:

I have nothing (no IDE) against which I can apply the patch, but what you did looks pretty good. If you want, I can upload files by which the deviation can be seen, so you can check it yourself.

Regarding 3m or 8m - its about 2.88m to be more exact. Is it possible that the deviation in the center is 0m and intensifies towards the edge ?

[Even Rouault]

You could just edit frmts/mbtiles/mbtilesdataset.cpp with any text editor. The patch means removing the line with #define MAX_GM 20037500. and replacing it with #define MAX_GM 20037508.3427892. Anyway, the shift should be uniform. And you should likely also see a shift in eastings. So the issue might be something else...

[ottoxt]

I have such a file is not on my computer. Do you know where I can change anything on a computer with Win7 and OSGeo4W?

[Even Rouault]

Replying to ottoxt:

I have such a file is not on my computer. Do you know where I can change anything on a computer with Win7 and OSGeo4W?

I assumed you built GDAL from source, but apparently not. Hum, no easy way to test the change then... and I'm a bit reluctant committing it without being sure it is appropriate.

[ottoxt]

Would it help if I put the mbtiles file from the attached jpg available (66m)? Or do you have a different suggestion?

[Even Rouault]

The geotiff and the mbtiles could help. With such sizes, you can't use a ticket attachment and shoud use instead an external hosting solution, dropbox, etc...

[ottoxt]

Here you go: ​https://www.dropbox.com/s/600i4rfohckei22/thetiff.zip?dl=0

[Even Rouault]

trunk r28884 "MBTiles: use standard EPSG:3857 origin to fix a ~8m shift (#5785)"

[Even Rouault]

trunk r28885 "Correctly update mbtiles.py expected result (#5785)"