[gdal-dev] Possible bug in GDAL 2.1 RC4 with MBTiles creation

[Even Rouault]

Le samedi 30 avril 2016 22:52:00, Gunnar Blom a écrit :
> Hi,
>  
> I've found a possible bug in GDAL 2.1 RC4 and the MBTiles driver.
>  
> I have the following image that I'm geo
> referencing: http://imgbox.com/tLFn3Auq 
> If I'm using the follwing code to warp the image based on a
> GeoTransform, it works
> well: https://gist.github.com/gunnarblom/9d73aa4358f3dafe2fd6808f7d4e9f41
> ...and it produces the following image which is
> correct: http://imgbox.com/TrzlXk4P 
> (note: as the target coordinate system is Apple's own in MapKit, the y-axis
> is reversed and therefor the negation of the two last parameters in the
> GeoTransform before warping) 
> If I'm using the following code to
> create MBTiles: https://gist.github.com/gunnarblom/76364b6bc8c0ff48dcf9fb8
> 1a5fdc060 ...but for some images/GeoTransforms it produces
> artifacts: http://imgbox.com/Gd15nfaP (three artifacts along the sides) 
> (note: this GeoTransform is based on WGS84-coordinates instead of Apple's
> coordinate system above in the regular image warp) 
> This error can be very strong on other images, it seems stronger the more
> skewed the transform is (example with a dummy-test I did with a random
> image: http://imgbox.com/FAWJj0eC) 
> What do you think can cause this bug?

I'm not sure. Does changing the GDAL_CACHEMAX configuration option changes the 
result (value in megabytes. default value is 5% of the total RAM) ?
Can you reproduce with gdal_translate ?

>  
> Thanks!
>  
> Regards,

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com