autocontext_mr_ct_model_zoo.md

Training regression network with autocontext

Downloading model zoo files

The training data and initial maps can be downloaded with the command

net_download autocontext_mr_ct_model_zoo

(Replace net_download with python net_download.py if you cloned the NiftyNet repository.)

Initial training

Command line parameters: --starting_iter 1 --max_iter 500

python net_regress.py train \
  -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
  --starting_iter 0 --max_iter 500

Generating contexts

Command line parameters: --spatial_window_size 240,240,1 --batch_size 4 modify the inference batch size and window size for efficiency purpose. The contexts will be generated to ~/niftynet/models/autocontext_mr_ct/autocontext_output.

python net_regress.py inference \
  -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
  --inference_iter 500 --spatial_window_size 240,240,1 --batch_size 4 --dataset_split_file nofile

Continue training with the latest contexts:

Command line parameters --starting_iter -1 indicate training the model from the most recently saved checkpoint (at iteration 500).

python net_regress.py train \
  -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
  --starting_iter -1 --max_iter 1000

Combine them together

Alternating in between context generation and training: (from git cloned source code)

python net_download.py autocontext_mr_ct_model_zoo
python net_regress.py train \
  -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
  --starting_iter 0 --max_iter 500
for max_iter in `seq 1000 1000 10000`
do
  python net_regress.py inference \
    -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
    --inference_iter -1 --spatial_window_size 240,240,1 --batch_size 4 --dataset_split_file nofile

  python net_regress.py train \
    -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \
    --starting_iter -1 --max_iter $max_iter
done

This script runs training for 10000 iterations, and new training contexts are generated at every 1000 iterations.

To see the training/validation curves using tensorboard:

tensorboard --logdir ~/niftynet/models/autocontext_mr_ct/logs

Generating regression output

Finally regression maps could be found at ~/niftynet/models/autocontext_mr_ct/autocontext_output/.

To make inferences using the sequence of trained models:

# reset the initial estimation folder:
net_download autocontext_mr_ct_model_zoo -r 
python net_regress.py inference \                                                                                                                                                                                                                 
    -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \                                                                                                                                                                              
    --inference_iter 500 --spatial_window_size 240,240,1 --batch_size 4 \                                                                                                                                                                         
    --dataset_split_file nofile                                                                                                                                                                                                                   
                                                                                                                                                                                                                                                  
for max_iter in `seq 1000 1000 10000`                                                                                                                                                                                                              
do                                                                                                                                                                                                                                                
    python net_regress.py inference \                                                                                                                                                                                                             
        -c ~/niftynet/extensions/autocontext_mr_ct/net_autocontext.ini \                                                                                                                                                                          
        --inference_iter $max_iter --spatial_window_size 240,240,1 --batch_size 4 \                                                                                                                                                               
        --dataset_split_file nofile                                                                                                                                                                                                               
done