for k, v in vars(args).items():
print(type(v), v)
if v is not None and hasattr(conf, k):
setattr(conf, k, v)
conf.reset_attributes()
conf.display()
with open(os.path.join(conf.test_dir, 'filenames.txt'), 'r') as reader:
test_files = reader.readlines()
test_files = [f.strip() for f in test_files]
test_gen = utils.BinaryData(conf.test_dir,
test_files,
conf.batch_size,
False,
dim=conf.dim)
reset_tf_session()
model = segnet.segnetCustomized((conf.dim, conf.dim, conf.input_channels),
1,
kernel=3,
pool_size=(2, 2),
output_mode=conf.final_activation)
h5_file = 'files/{}_{}_{}_{}.h5'.format(conf.specific_name,
conf.model_type, conf.dim,
conf.epochs)
opt = keras.optimizers.Adam(0.001)
conf.reset_attributes()
conf.display()
with open(os.path.join(train_dir, 'filenames.txt'), 'r') as reader:
train_files = reader.readlines()
train_files = [f.strip() for f in train_files]
with open(os.path.join(valid_dir, 'filenames.txt'), 'r') as reader:
valid_files = reader.readlines()
valid_files = [f.strip() for f in valid_files]
train_gen = utils.BinaryData(conf.train_dir,
train_files,
conf.batch_size,
True,
dim=conf.dim)
valid_gen = utils.BinaryData(conf.valid_dir,
valid_files,
conf.batch_size,
False,
dim=conf.dim)
reset_tf_session()
model = segnet.segnetCustomized((conf.dim, conf.dim, conf.input_channels),
1,
kernel=3,
pool_size=(2, 2),
output_mode=conf.final_activation)
help='crop labels to half or not',
type=bool)
args = parser.parse_args()
for k,v in vars(args).items():
print(type(v), v)
if v is not None and hasattr(conf, k):
setattr(conf, k, v)
conf.reset_attributes()
conf.display()
with open(os.path.join(conf.test_dir, 'filenames.txt'), 'r') as reader:
test_files = reader.readlines()
test_files = [f.strip() for f in test_files]
test_gen = utils.BinaryData(conf.test_dir, test_files, conf.batch_size, False, dim=conf.dim, return_filenames=True)
keras.backend.clear_session()
model = segnet.segnetCustomized((conf.dim, conf.dim, conf.input_channels), 1, kernel=3, pool_size=(2, 2), output_mode=conf.final_activation)
h5_file = 'files/{}_{}_{}_{}.h5'.format(conf.specific_name, conf.model_type, conf.dim, conf.epochs)
opt = keras.optimizers.Adam(0.001)
metrics = ['acc', utils.f1_m, utils.precision_m, utils.recall_m]
model.load_weights(h5_file)
model.compile(loss='binary_crossentropy', optimizer=opt, metrics=metrics)
output_shp_file = 'files/{}'.format(conf.output_shp_file)
utils.make_predictions_and_vectorize_binary(model, test_gen, output_shp_file, conf.test_dir)
# utils.make_predictions_and_vectorize(model, test_gen, output_shp_file)