return_01c=False,
return_list=False,
use_threads=use_threads,
preload=True,
n_layers=n_layers)
n_batches_train = train_iter.nbatches
n_batches_val = val_iter.nbatches
n_classes = 2
void_labels = train_iter.void_labels
# Define model
simple_net_output, net = build_simple_model(input_var,
filter_size=filter_size,
n_filters=n_filters,
depth=depth,
block=block,
nb_in_channels=nb_in_channels,
n_classes=n_classes)
# load weights to balance training error
Class = np.loadtxt(
'/Tmp/cucurulg/datasets/cortical_layers/6layers_segmentation/training_cls.txt'
)
def compute_class_weights(Class):
#get unique labels and number of pixels of each class
unique, counts = np.unique(Class, return_counts=True)
#calculate freq(c) number of pixels per class divided by the total number of pixels in images where c is present
freq = counts.astype(float) / Class.size
shuffle_at_each_epoch = True,
return_one_hot=False,
return_01c=False,
return_list=False,
use_threads=use_threads,
preload=True,
n_layers=args['n_layers'])
n_batches_val = val_iter.nbatches
n_classes = val_iter.non_void_nclasses
void_labels = val_iter.void_labels
simple_net_output, net = build_simple_model(input_var,
filter_size = args['filter_size'],
n_filters = args['n_filters'],
depth = args['depth'],
block= args['block'],
nb_in_channels = nb_in_channels,
n_classes = n_classes)
Class=np.loadtxt('/Tmp/cucurulg/datasets/cortical_layers/6layers_segmentation/training_cls.txt')
def compute_class_weights(Class):
#get unique labels and number of pixels of each class
unique, counts = np.unique(Class,return_counts=True)
#calculate freq(c) number of pixels per class divided by the total number of pixels in images where c is present
freq=counts.astype(float)/Class.size
return np.median(freq)/freq
weights=compute_class_weights(Class)
print "Defining and compiling valid functions"