[MINC-development] Grid transforms

[Andrew Janke]

Yup, this does work, (along with using anything other than a signed type).

The problem is that this only goes awry when you pass the file as part
of an xfm to xfmconcat, not during routine mincreshape/mincmath games.

So, I'd argue it's a bug in volume_io (concatenate_general_transforms)
but it comes down to what is defined behaviour in MINC with grid
transforms. As far as I am aware this isn't defined and largely
"whatever minctracc puts out".

If this is the case should we:

a) Warn the user if an input grid file is given which isn't either
unsigned short or double (currently used options by minctracc)

b) Convert it to something that will work.


ta

a



On 22 December 2016 at 11:54, Vladimir S. FONOV
<vladimir.fonov at gmail.com> wrote:
> Try specifying -image_range 0 1 -valid_range 0 1 in mincreshape
>
> On Dec 21, 2016 8:47 PM, "Andrew Janke" <a.janke at gmail.com> wrote:
>>
>> Hi all,
>>
>> I understand that a while back the precision of grid transforms went
>> from short to double. This was probably around the -50 -> + 50
>> wierdness that used to happen within minctracc.  Perhaps there were
>> also other rationales.
>>
>> Were there also other changes to libminc/volume_io to support this? I
>> ask as I'm getting wierdness with xfmconcat when I feed in a zero grid
>> transform.  ie:
>>
>> mincresample -xstep 32 -ystep 32 -zstep 32 \
>>    -xnelements 10 -ynelements 10 -znelements 10 \
>>    /opt/minc/share/icbm152_model_09c/mni_icbm152_t1_tal_nlin_sym_09c.mnc
>> /tmp/like.mnc
>>
>> minclookup -signed -short -lut_string '0 0 0 0;1 0 0 0'  /tmp/like.mnc
>> /tmp/like_vector.mnc
>>
>> mincreshape -dimorder vector_dimension,zspace,yspace,xspace \
>>    /tmp/like_vector.mnc /tmp/zero_grid_0.mnc
>>
>> echo "MNI Transform File" > /tmp/zero.xfm
>> echo "Transform_Type = Grid_Transform;" >> /tmp/zero.xfm
>> echo "Displacement_Volume = zero_grid_0.mnc;" >> /tmp/zero.xfm
>>
>> xfmconcat /tmp/zero.xfm /tmp/z.xfm
>>
>> mincstats /tmp/z.xfm
>>
>> cai-zeep:tmp$ mincstats z_grid_0.mnc
>> *** mincstats - reported max (-32768) doesn't equal header (32767)
>> File:              z_grid_0.mnc
>> Mask file:         (null)
>> Total voxels:      3000
>> # voxels:          3000
>> % of total:        100
>> Volume (mm3):      98304000
>> Min:               -32768
>> Max:               -32768
>>
>> Note the messed up min and max. This will work however if you make an
>> unsigned short volume at the minclookup stage.   If I then use this
>> transform things of course go awry instead of it being a zero xfm.
>>
>> So, is this defined or unexpected behaviour? If I use a double, I get
>> a more interesting error instead of a failure.
>>
>> xfmconcat ident-32.xfm /tmp/d.xfm
>> Volume has invalid min and max voxel value
>>
>>
>> ta
>>
>>
>> a
>> _______________________________________________
>> MINC-development mailing list
>> MINC-development at bic.mni.mcgill.ca
>> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-development
>
>
> _______________________________________________
> MINC-development mailing list
> MINC-development at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-development
>