Disagreement between miget_real_value and miget_real_value_hyperslab #108
Description
Typically in RMINC we read data in using miget_real_value_hyperslab. However one of our functions, mincGetVoxel uses miget_real_value. I would have assumed that these two functions would agree on any particular voxel's real value, but that doesn't seem to be the case at least for double precision volumes. Internally miget_real_value uses mirw_hyperslab_raw which does no conversion on the value, and then post-hoc calls miconvert_voxel_to_real. Whereas miget_real_value_hyperslab calls mirw_hyperslab_icv which does its own, apparently more complicated scaling internally. So I suspect that miconvert_voxel_to_real is missing something relevant for the rescaling.
Here's a sample C program illustrating the problem
test-miget-real-voxel.c
#include <minc2.h>
#include <stdio.h>
int main(int argc, char** argv){
char* file=argv[1];
misize_t location[3];
misize_t count[3];
mihandle_t hvol;
int result;
int i;
double single;
double slab;
for(i = 0; i < 3; i++){
location[i] = 50;
count[i] = 1;
}
result = miopen_volume(file,
MI2_OPEN_READ, &hvol);
if (result != MI_NOERROR) {
printf("Error opening input file: %s.\n", file);
return 1;
}
result = miget_real_value(hvol, location, 3, &single);
printf("Getting a single value returns: %f.\n", single);
result = miget_real_value_hyperslab(hvol, MI_TYPE_DOUBLE, location, count, &slab);
printf("Getting a size 1 slab returns: %f.\n", slab);
return 0;
}An easy way to test is to convert the icbm average to unsigned double
mincresample -2 -double mni_icbm152_t1_tal_nlin_sym_09a.mnc mni_icbm152_t1_tal_nlin_sym_09a_signed_double.mnc
Then compile and run the above program to check the difference, on my machine I get
Getting a single value returns: 15.561608.
Getting a size 1 slab returns: 17.435906.
This was noticed by one of our users getting the expected answer when using mincAnova which uses miget_real_value_hyperslab internally, but a very different answer when plucking individual voxels with mincGetVoxel which uses miget_real_value