def generate_images(path,
new_foldername="images",
post_processing=False,
results_fname="test_results.hdf5"):
""" Vygeneruje obrazky pro porovnani """
images_path = path + "/" + new_foldername
fm.make_folder(images_path)
file = h5py.File(path + "/" + results_fname, 'r')
test_data = file["test_data"]
test_labels = file["test_labels"]
test_predictions = file["test_predictions"]
# dataname = fm.load_json(path+"/notebook_config.json")["experiment_name"]
# file2 = h5py.File("datasets/processed/"+dataname+".hdf5", 'r')
# test_data = file2["test_data"]
for index in range(test_data.shape[0]):
if index % 100 == 0:
print("Generuji se obrazky:", index, "/", test_data.shape[0])
orig = test_data[index][:, :, 0].astype("uint8")
label = test_labels[index].astype("uint8") * 255
result = test_predictions[index].astype("float")
lesion = np.argmax(result, axis=2) * 127
if post_processing:
post, _ = CNN_evaluator.apply_morphology_operations(lesion, label)
post[lesion != 0] += 1
blank = np.zeros(label.shape, dtype="uint8")
blank[post[:] == 0] = [255, 0, 0]
blank[post[:] == 1] = [0, 0, 255]
blank[post[:] == 2] = [0, 255, 0]
show_results(
label,
orig,
blank,
result,
lesion_label=u"Výstup CNN",
result_label=u"Výstup CNN po morfologických operacích",
path=images_path,
index=index)
else:
show_results(label,
orig,
result,
lesion,
path=images_path,
index=index)
#break
file.close()
loss = args.loss
LR_change = args.LR_change
batch_size = args.batch_size
special_label = optimizer_label + "_SegNetSkip_Morph"
special_label = special_label + "_LR" + str(abs(int(np.log10(LR)))) + LR_change
special_label = special_label + "_" + str(epochs) + "epochs"
special_label = special_label + "_" + str(batch_size) + "bs"
""" Nacteni dat """
#experiment_name = "no_aug_structured_data-liver_only"
experiment_name = "aug_structured_data-liver_only"
#experiment_name = "aug-ge+int_structured_data-liver_only"
experiment_foldername = "experiments/" + experiment_name
fm.make_folder(experiment_foldername)
hdf_filename = "datasets/processed/" + experiment_name + ".hdf5"
hdf_file = h5py.File(hdf_filename, 'r')
train_data = hdf_file['train_data'][:]
train_labels = hdf_file["train_labels"][:]
val_data = hdf_file['val_data'][:]
val_labels = hdf_file["val_labels"][:]
""" Architektura """
h, w, d = train_data.shape[1:]
inputs = Input(shape=(h, w, d))
# downsampling
xc1 = Conv2D(64, (3, 3), padding='same', activation='relu',
strides=(1, 1))(inputs)
if addr.startswith(name):
if name in train_names:
train_addrs.append(addr)
elif name in val_names:
val_addrs.append(addr)
elif name in test_names:
test_addrs.append(addr)
print(len(train_addrs), len(val_addrs), len(test_addrs))
return train_addrs, val_addrs, test_addrs
""" Definice cest """
fm.make_folder("datasets/processed/")
hdf_path = "datasets/processed/aug_structured_data-liver_only.hdf5"
data_path = 'CTs/kerasdata/Slices/*.png'
masks_path = 'CTs/kerasdata/Masks/*.png'
data_addrs = glob.glob(data_path)
masks_addrs = glob.glob(masks_path)
# jde to vzit i bez augmentace
data_addrs = [a for a in data_addrs]# if "rot=0_shear=0" in a]
masks_addrs = [a for a in masks_addrs]# if "rot=0_shear=0" in a]
print(len(data_addrs), len(masks_addrs))
@author: Mirab
"""
import glob
import os
import re
from shutil import copyfile
import file_manager_metacentrum as fm
foldername = "experiments/aug_structured_data-liver_only/*"
folders = glob.glob(foldername)
for folder in folders:
name = re.sub(r".+\\+", "", folder)
if "." in name:
continue
new_name = "experiments_results_copy/"
fm.make_folder(new_name + name + "/logs")
files = os.listdir(folder)
for f in files:
if f.endswith(".json"):
copyfile(folder + "/" + f, new_name + name + "/" + f)
if f.endswith("logs"):
logfiles = os.listdir(folder + "/" + f)
for lf in logfiles:
if not (lf.endswith(".hdf5") or lf.endswith(".h5")):
copyfile(folder + "/" + f + "/" + lf,
new_name + name + "/" + f + "/" + lf)