How to specify input/output pairs in /images/list.txt for model training

[adityaapte]

Task (what are you trying to do/register?)

Train a model from pairs of registered CT images (3D).

What have you tried

I am at data preparation stage in the format required by training script, but not sure how files names for input/output image pairs should be specified in /images/list.txt.

Details of experiments

I haven't run training yet.

[adalca]

I believe just one filename per line in list.txt, and (if you dont provide an atlas) the script will randomly choose pairs.

if you already have predefined pair, you can quickly modify the script.py file here:

voxelmorph/scripts/torch/train.py

Line 97 in ca28315

train_files = vxm.py.utils.read_file_list(args.img_list, prefix=args.img_prefix,

to use read_pair_list instead of read_file_list.

[adityaapte]

Thanks!