Skip to content

Latest commit

 

History

History
35 lines (30 loc) · 2.44 KB

README.md

File metadata and controls

35 lines (30 loc) · 2.44 KB
Raw

run scripts

Aside from the jupyter notebooks showcased as examples we provide these scripts to be able to run each step directly from the command line by just providing an appropriate configuration file

  • 01_train.py Run the training process. The configuration file determines, among other things, what data should be used for training (train_data), what kind of model should be used (model) and for how long it should be trained (train) and with what kind of augmentations (train.augment).
  • 02_predict.py Run the prediction process. The configuration file determines which previously trained model should be used for prediction and what data should be used (validate_data, test_data).
  • 03_extract_edges.py Run the extract edges process (For each detected cell candidate look in the spatial neighborhood of the previous time frame to find potential parent cell candidates). The configuration file determines how far a cell could have moved at most to count as a potential link.
  • 04_solve.py Run the solve process (tracking). The configuration file determines what ILP weights should be used to solve the ILP, and if a grid search should be executed to find optimal weights.
  • 05_evaluate.py Run the evaluation process. The configuration file determines how close detected cell and ground truth annotation have to be to be counted as a correct match and what ROI (region of interest) of the provided data we want to evaluate on.
  • 06_run_best_config.py Run solve and evaluate on the test data using the best parameters/weights as determined on the validation data.

For further convenience we provide a final script that, depending on the given flags, calls the appropriate script with the correct arguments: linajea (should be called from within an experiment directory)

mkdir $setup_dir
cd $setup_dir
linajea --config config.toml --train

For information on the available flags use python linajea --help:

usage: linajea [-h] [--config CONFIG] [--checkpoint CHECKPOINT]
               [--train] [--predict] [--extract_edges] [--solve] [--evaluate] [--best]
               [--validation] [--validate_on_train]
               [--param_id PARAM_ID] [--val_param_id VAL_PARAM_ID] [--param_ids PARAM_IDS [PARAM_IDS ...]]
               [--local] [--slurm] [--gridengine] [--interactive]
               [--array_job] [--eval_array_job] [--wait_job_id WAIT_JOB_ID] [--no_block_after_eval]