import tensorflow as tf
import keras
from keras.backend.tensorflow_backend import set_session
from keras.optimizers import Adam, Adagrad
from keras.callbacks import ModelCheckpoint, Callback, History
from dl_utils.dl_networks.resnet import ResnetBuilder
from dl_utils.tb_callback import TensorBoard
K.set_image_dim_ordering('th')
##############################
#
# Config data
#
##############################
cases = dataset_loaders.get_casenames(test=True)
image_size_nn = 45
N_valid_cases = 50
wind = 100
big_patch_size = 100
experiment_folder_name = 'final_classificator'
experiment_name = 'final_classificator' # This one will only be used for the logs
INPUT_MODEL = '%s/%s/models/final_classificator_model_2.hdf5' % (
settings.DATAMODEL_PATH, experiment_folder_name)
existing_labels = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'FP']
##############################
scan_step,
batch_size,
square_to_scan=None)
preds = []
for x in patches:
preds.append(model.predict(x))
preds = np.vstack(preds)
preds = preds.reshape([
int(np.sqrt(preds.shape[0])),
int(np.sqrt(preds.shape[0])), preds.shape[1]
])
return preds, time.time() - t1
for test in [False, True]:
for casename in dataset_loaders.get_casenames(test=test):
if casename == 'emtpy':
continue
try:
filename_to_save = annotations_name.format(path=ANNOTATIONS_PATH,
scan_window=scan_step,
casename=casename)
if not os.path.isfile(filename_to_save):
preds, takentime = predict_case(casename)
print(
"- Calculated case %s with window step %d in %0.0f seconds"
% (casename, scan_step, takentime))
np.savez_compressed(filename_to_save, preds=preds)
else:
print("- [ALREADY DONE] %s" % filename_to_save)
except:
def predict_case(case):
t1 = time.time()
patches = scan_patches(case,
image_size_nn,
patch_size,
scan_step,
batch_size,
square_to_scan=None)
preds = []
for x in patches:
preds.append(model.predict(x))
preds = np.vstack(preds)
preds = preds.reshape([
int(np.sqrt(preds.shape[0])),
int(np.sqrt(preds.shape[0])), preds.shape[1]
])
return preds, time.time() - t1
for casename in dataset_loaders.get_casenames():
filename_to_save = annotations_name.format(path=ANNOTATIONS_PATH,
scan_window=scan_step,
casename=casename)
if not os.path.isfile(filename_to_save):
preds, takentime = predict_case(casename)
print("- Calculated case %s with window step %d in %0.0f seconds" %
(casename, scan_step, takentime))
np.savez_compressed(filename_to_save, preds=preds)
else:
print("- [ALREADY DONE] %s" % filename_to_save)