'mvt_train_cati_ms'
]
nb_motions_list = [20, 5, 5, 10, 10, 5, 5]
nb_motions_list = [20, 20, 20, 10, 10, 10] #[5, 5, 5]
#nb_motions_list = [50, 50, 50, 5, 5, 5]
#nb_motions_list = [50, 50, 50, 20, 20, 20]
#nb_motions_list = [200]
nb_motions_list = [20, 5, 5, 10, 10, 5, 5]
name_list = ['mvt_val_cati_T1', 'mvt_val_cati_ms']
nb_motions_list = [20, 20]
do_plotting = False
fin_list_train, fin_list_val = get_train_and_val_csv(name_list) #
fin_choose = []
for ii, nn in enumerate(name_list):
if 'train' in nn: fin_choose.append(fin_list_train[ii])
elif 'val' in nn: fin_choose.append(fin_list_val[ii])
fin_list = [pd.read_csv(ff).filename for ff in fin_choose]
for name, fin, nb_motions in zip(name_list, fin_list, nb_motions_list):
resdir = prefix + name + '/'
scriptsDir = '/network/lustre/iss01/cenir/software/irm/toolbox_python/romain/torchQC'
if 'mvt' in name:
motion_type = 'motion1'
data_name_train = name_list_train[3]
data_name_val = name_list_val[3]
nb_replicate = 20
if do_eval:
nb_replicate = 3
res_dir = '/network/lustre/iss01/cenir/analyse/irm/users/romain.valabregue/QCcnn/NN_regres_random_noise/'
base_name = 'Reg_AffN'
if make_uniform: base_name += '_uniform'
root_fs = 'le70' #
#root_fs = 'lustre'
train_csv_file, val_csv_file = get_train_and_val_csv(data_name_train,
root_fs=root_fs)
if do_eval:
train_csv_file = val_csv_file
par_model = {
'network_name': 'ConvN',
'losstype': 'L1',
'lr': 1e-4,
'conv_block': [16, 32, 64, 128, 256],
'linear_block': [40, 50],
'dropout': 0,
'batch_norm': True,
'drop_conv': 0.1,
'validation_droupout': False,
'in_size': in_size,