Skip to content

Latest commit

 

History

History
24 lines (21 loc) · 1.07 KB

inference.md

File metadata and controls

24 lines (21 loc) · 1.07 KB

Running inference with a pre-trained DEF network

For the most part, running testing with a pre-trained DEF model is described in training manual.

Here, we mention a helper SLURM script run_inference.sbatch.shhttps://github.com/artonson/def/blob/main/sharpf/neural/run_inference.sbatch.sh which serves as an interface to the inference script. You can use the script as an example for building you own inference script.

The output structure for the inference looks like this:

(base) [a.artemov@an01:/gpfs/gpfs0/3ddl/sharp_features/predictions/images_align4mm_fullmesh_whole/amed/92side_folder_images__align4mm_fullmesh_whole]$ll
total 3.5K
drwxrwsr-x  6 3ddl 3ddl 4.0K May  6  2021 .
drwxrwsr-x 56 3ddl 3ddl 4.0K May 17  2021 ..
drwxrwsr-x  3 3ddl 3ddl 4.0K May  6  2021 default
drwxrwsr-x  2 3ddl 3ddl 4.0K May  6  2021 .hydra
drwxrwsr-x  2 3ddl 3ddl 4.0K May  6  2021 predictions
drwxrwsr-x  3 3ddl 3ddl 4.0K May  6  2021 tb_logs
-rwxrwsr-x  1 3ddl 3ddl 1.1K May  6  2021 train_net.log

which is what is required for the fusion method to work.