bold_MNIInfant_wf uses res-1 (1mm) template as target for BOLD resampling

What happened?

We noticed that when using --cifti-output-91k, the bold_MNIInfant_wf resample node uses the res-1 (1mm) template as the reference when resampling the BOLD timeseries to MNIInfant space, instead of selecting the res-2 (2mm) template as in NiBabies versions prior to the fit-apply overhaul. This has been causing OOM errors for us and presumably is also adding unintended smoothing to the CIFTI BOLD from resampling it to 1mm and then back to 2mm in the final CIFTI output.

What command did you use?

/opt/conda/envs/nibabies/bin/nibabies /bids_dir /output_dir participant --participant-label XXXXXX -s 4mo --omp-nthreads 8 --nprocs 8 --cifti-output 91k -vv -w /wd --fd-radius 35 --derivatives /derivatives --age-months 4 --skip_bids_validation --surface-recon-method mcribs --project-goodvoxels

What version of NiBabies are you using?

24.0.0rc0

Relevant log output

(report.rst for bold_MNIInfant_wf resample node)
---


Node: single_subject_sub-XXXXXX_ses-Xmo_wf (bold_ses_Xmo_task_rest_acq_AP_run_02_wf (bold_MNIInfant_wf (resample (resampling)
=============================================================================================================================


 Hierarchy : nibabies_24_0_wf.single_subject_sub-XXXXXX_ses-Xmo_wf.bold_ses_Xmo_task_rest_acq_AP_run_02_wf.bold_MNIInfant_wf.resample
 Exec ID : resample


Original Inputs
---------------


* cval : 0.0
* fieldmap : <undefined>
* in_file : /wd/nibabies_24_0_wf/single_subject_sub-XXXXXX_ses-Xmo_wf/bold_ses_Xmo_task_rest_acq_AP_run_02_wf/bold_native_wf/bold_stc_wf/copy_xform/sub-XXXXXX_ses-Xmo_task-rest_acq-AP_run-02_bold_tshift_xform.nii.gz
* inverse : [False, False, False]
* jacobian : True
* mode : constant
* num_threads : 8
* order : 3
* output_data_type : float32
* pe_dir : <undefined>
* prefilter : True
* ref_file : /home/faird/lundq163/.cache/templateflow/tpl-MNIInfant/cohort-2/tpl-MNIInfant_cohort-2_res-1_desc-brain_mask.nii.gz
* ro_time : <undefined>
* transforms : ['/output_dir/sub-XXXXXX/ses-Xmo/func/sub-XXXXXX_ses-Xmo_task-rest_acq-AP_run-02_from-orig_to-boldref_mode-image_desc-hmc_xfm.txt', '/output_dir/sub-XXXXXX/ses-Xmo/func/sub-XXXXXX_ses-Xmo_task-rest_acq-AP_run-02_from-boldref_to-T2w_mode-image_desc-coreg_xfm.txt', '/wd/nibabies_24_0_wf/single_subject_sub-XXXXXX_ses-Xmo_wf/infant_anat_fit_wf/register_template_wf/_template_MNIInfantcohort-2/registration/ants_t1_to_mniComposite.h5']

Add any additional information or context about the problem here.

We tried a workaround of adding 'res': vol_res where the MNIInfant reference is added here:

nibabies/nibabies/workflows/base.py

Line 843 in fa07eaf

spaces.add(Reference('MNIInfant', {'cohort': cohort}))

, but ran into another error in select_MNIInfant_xfm since the res-2 is not included in the keys corresponding to the available transforms.