Trouble running dataset
savvvyk opened this issue · comments
Hello,
I am trying to run the 2016 FLI-AIM MSSEG processed data set through the nicMSlesion_linux tool. I have tried both using a manual setup of the requirements with the UI and the docker container. Both ways I run into the same issues when attempting to run the data (configuration below as well):
If I attempt using the pre-trained baseline_2ch configuration or if I attempt without any pre-trained models:
The terminal shows
CNN: loading training data for first model
and after ~5 minutes the UI crashes with no error messages showing
Configuration (shows when attempting all 4 channels without a pre-trained model in use):
[database]
train_folder = /home/MS_data/2016_fli_aim_msseg_MS_data_processed
inference_folder = /path/to/testing
flair_tags = FLAIR_preprocessed
t1_tags = T1_preprocessed
mod3_tags = T2_preprocessed
mod4_tags = DP_preprocessed
roi_tags = Consensus
register_modalities = False
denoise = False
denoise_iter = 3
skull_stripping = False
save_tmp = True
debug = True
[train]
full_train = True
pretrained_model = baseline_2ch
balanced_training = True
fraction_negatives = 2.0
[model]
name = MSSEG
pretrained = None
train_split = 0.25
max_epochs = 400
patience = 50
batch_size = 128
net_verbose = 1
gpu_number = 0
[postprocessing]
t_bin = 0.5
l_min = 10
min_error = 0.5
Any help would be appreciated.
Thanks,
savvvyk