can't malloc -1437173116 bytes in 3dNwarpApply

ruyuanzhang · November 21, 2018, 9:41pm

Hi, afni experts
I am trying to concatenate an affine (from 3dAllineate) and a nonlinear transform (from SSwarper) in 3dNwarpApply. However, the command keeps crashing. Actually, I tried to apply the affine and the nonlinear separately and both got good results. However, if I concatenate them. Commands crashed. Also, the exact same command worked on other subjects but kept crashing on one subject. Did I do something wrong? The error message is below.

ruyuan@stone:~/Dropbox/stonesync/monkeyanatpp/starlord$ 3dNwarpApply -prefix NMTseg_dti_FA.nii.gz -nwarp ‘dti_FA_affepi2t1_t1_matINV.aff12.1D starlord_anatT1_NMT2origWARP.nii.gz’ -source /home/stone/ruyuan/afniresource/NMT/NMT_v1.2/atlases/D99_atlas/D99_atlas_1.2a_al2NMT.nii.gz -master dti_FA.nii.gz -ainterp NN -overwrite -short
++ 3dNwarpApply: AFNI version=AFNI_18.2.19 (Sep 27 2018) [64-bit]
++ Authored by: Zhark the Warped
++ opened source dataset ‘/home/stone/ruyuan/afniresource/NMT/NMT_v1.2/atlases/D99_atlas/D99_atlas_1.2a_al2NMT.nii.gz’
++ -master dataset is ‘dti_FA.nii.gz’
++ Processing -nwarp
++ Warping:** ERROR: EDIT_full_copy: can’t malloc -1437173116 bytes for new sub-brick 0

Dataset ./Nwarp#02+orig.HEAD

Fatal Signal 11 (SIGSEGV) received
THD_nwarp_regrid
IW3D_set_geometry_nwarp_catlist
IW3D_read_nwarp_catlist
3dNwarpApply
Bottom of Debug Stack
** Command line was:
3dNwarpApply -prefix NMTseg_dti_FA.nii.gz -nwarp ‘dti_FA_affepi2t1_t1_matINV.aff12.1D starlord_anatT1_NMT2origWARP.nii.gz’ -source /home/stone/ruyuan/afniresource/NMT/NMT_v1.2/atlases/D99_atlas/D99_atlas_1.2a_al2NMT.nii.gz -master dti_FA.nii.gz -ainterp NN -short
** AFNI version = AFNI_18.2.19 Compile date = Sep 27 2018
** [[Precompiled binary linux_openmp_64: Sep 27 2018]]
** Program Death **
** If you report this crash to the AFNI message board,
** please copy the error messages EXACTLY, and give
** the command line you used to run the program, and
** any other information needed to repeat the problem.
** You may later be asked to upload data to help debug.
** Crash log is appended to file /home/stone/ruyuan/.afni.crashlog

Best
ruyuan

ruyuanzhang · November 21, 2018, 10:07pm

Add another piece of information here. I made two further explorations:

I used 3dNwarpCat to combine the affine and the nonlinear warp file. Then use 3dNwarpApply to apply the obtained warp file. It reported no crash but produced very bad warp results.
I ran two stages transformation. First did the nonlinear warp, saved the result. Then ran the affine. The results looked reasonable. However I have to resample twice in this regime.

It seems that 3dNwarpCat did not very well…

ptaylor · November 21, 2018, 11:14pm

Hi, Ruyuan-

First question: @SSwarper on its own produces both an affine (*aff12.1D) and a nonlinear (WARP) transform that should be concatenated-- I’m not sure why a separate call to 3dAllineate would be needed or desired?

Also, from the 3dNwarpApply help file, I think the order of your concatenation in your original command is backwards:


3dNwarpApply -prefix Fred_final    \
                     -source Fred+orig     \
                     -master NWARP         \
                     -nwarp 'Fred_WARP+tlrc Fred.Xaff12.1D'

–pt

ruyuanzhang · November 26, 2018, 10:50pm

Thanks Paul

I actually first align my macaque T1 image to the NMT macaque template and obtained a nonlinear transformation file (i.e., WARP.nii.gz). I checked that this transformation was correct. I then align my EPI image to T1 and obtained an affine transformation matrix (i.e.,_mat.aff12.1D). My goal is to extract an ROI defined in the NMT template and warp back to the EPI space. In this case, I have to concatenate the affine the and nonlinear transformation matrix.

I did this for other three monkeys and it worked very well. It however always made the program crashed on one monkey brain. I am very confused. Now I am not able to concatenate them and have to resample twice, NMT template - t1 - epi.

I can provide the data if you are willing to take a look.

ruyuan

dglen · November 26, 2018, 10:57pm

The way the transformations are concatenated can take a lot of memory if the datasets are not close in space to each other. I typically will use @Align_Centers to put the center of the dataset over the template’s center. The NMT align script and the macaque_align.csh script do this first before computing alignment. The SSwarper script requires a specially crafted dataset to do its business.

ruyuanzhang · November 27, 2018, 8:41pm

Thanks Daniel. I think that might be the reason. I indeed did the @align_center step when I aligned data but combined the shift matrix to other affine transformations. If memory is an issue, I need to keep the shift matrix.

Thanks again

ruyuan

dglen · November 27, 2018, 10:05pm

That needs to be separate if you can manage it. Usually, you can work around this issue because there is no interpolation done by @Align_Centers because that transformns by just changing the origin in the header using 3drefit. The center shift transformation can be applied to other datasets with the -child option for @Align_Centers. If you are interested in the inverse transformation, you can apply a transformation via @Align_Centers again with the base and dset datasets reversed.