The problem is that estimate procedure uses the same library (at least
gives the same error) so I'll have to move my whole pipeline to
multiple 3D files. Back to gazzillion files using up my inodes ;) Out
of curiosity: are you planning to use only nifti/file_array in the
future?
Thanks,
Chris
2010/1/22 Guillaume Flandin <[log in to unmask]>:
> Dear Chris,
>
> sorry, the error message you get is a bit unfortunate when you are
> actually dealing with a large file...
> The problem happens because SPM has concurrently two libraries to access
> data files (nifti/file_array and spm_mapping): the former has Large File
> Support (LFS) while the latter hasn't and that's the one that the
> smoothing operation uses...
> So for now my suggestion would be to convert your 4D file into many 3D
> files (using spm_file_split) and smooth those. If you really want a 4D
> file at the end of the day, you can re-assemble them afterwards with
> spm_file_merge.
>
> Best regards,
> Guillaume.
>
>
> Chris Filo Gorgolewski wrote:
>> Hi,
>> I am trying to smooth a big 4D .nii file (7.5GB, 256x176x200x446). The
>> smoothing function gives me this error:
>>
>> [...]
>>
>> ------------------------------------------------------------------------
>> Running job #1
>> ------------------------------------------------------------------------
>> Running 'Smooth'
>> Failed 'Smooth'
>> Error using ==> spm_conv_vol
>> File too small.
>> In file "/home/filo/opt/spm8/spm_smooth.m" (v2794), function "smooth1"
>> at line 105.
>> In file "/home/filo/opt/spm8/spm_smooth.m" (v2794), function
>> "spm_smooth" at line 37.
>> In file "/home/filo/opt/spm8/config/spm_run_smooth.m" (v3534),
>> function "spm_run_smooth" at line 20.
>>
>> I have further located the error message to be in get_map_file from
>> spm_mapping.c. Smooth job before crashing creates a file with the
>> first 120 smoothed volumes (2gb in size). I have ruled out memory
>> allocation problems by running it on 32gb machine.
>>
>> Its a 64bit linux machine with spm8 r3684. Does anyone have a clue
>> what might be wrong?
>>
>> Best regards,
>> Chris
>>
>>
>
>
> --
> Guillaume Flandin, PhD
> Wellcome Trust Centre for Neuroimaging
> University College London
> 12 Queen Square
> London WC1N 3BG
>
|