[gdal-dev] Bigger file size from 64-bit than 32-bit

[Even Rouault]

Selon Jukka Rahkonen <jukka.rahkonen at mmmtike.fi>:

> Even Rouault <even.rouault <at> mines-paris.org> writes:
>
> > --> ouch ! I fail to see any reason why the on-disk format would be
> affected by
> > 32 vs 64 bit. You could run "tiffdump" on the generated images to see the
> offset
> > and size of the tiles. What is surprising is that the 32 bit JPEG TIF is
> smaller
> > than the original JP2. Does it display correctly ?
> > Does gdalinfo -checksum return the same values ?
>
> Ok, the images in this B/W set converted with 64-bit binaries are not only
> big but they are totally broken. I will need to check the B/W images I have
> processed so far with these binaries and see if they are all broken. I know
> that colour images are good and I believe that I have also checked a bunch
> of B/W images. This certain B/W batch is compressed into JPEG2000 a long
> time ago with another program and settings than I use nowadays so it can be
> that other images are not affected. I will tell more after doing some
> inspections.
>
> Originals are compressed as lossless JPEG2000 so it is normal that a jpeg
> compressed tiff is much smaller.

What should be determined is if the problem is on the read side (JPEG 2000
decoding) or at the write side (JPEG in TIFF compression). Perhaps you could
translate to uncompressed TIFF. My feeling is that the problem must be on the
read side and that you get more or less random images at each read attempt. You
could also run gdalinfo -checksum on the JPEG2000 image itself and see if it
returns a constant value.