def test(self, x_test, y_test, data_start, batch_size=10, threshold=0.5, save_path = None):
def mkdir(path):
# 去除首位空格
path = path.strip()
# 去除尾部 \ 符號
path = path.rstrip("\\")
# 判斷路徑是否存在
# 存在 True
# 不存在 False
isExists = os.path.exists(path)
# 判斷結果
if not isExists:
# 如果不存在則建立目錄
print("Building the file.")
# 建立目錄操作函式
os.makedirs(path)
return True
else:
# 如果目錄存在則不建立，並提示目錄已存在
print("File is existing.")
return False
# if x_test.ndim == 3:
# x_test = np.expand_dims(x_test, axis=-1)
# if y_test.ndim == 3:
# y_test = np.expand_dims(y_test, axis=-1)
if save_path == None:
save_path = self.name
# with open(".\\result\\" + self.name + 'estimator_weights.json', 'r', encoding='utf-8') as f:
# output = json.load(f)
# estimator_weights = output['estimator_weights']
file_name = []
for order in range(self.n_estimators):
file_name.append(self.name + "_" + str(order) + ".h5")
print(file_name[-1])
y_predict = np.zeros(y_test.shape, dtype=np.float32)
c = 0
for model_file in file_name:
# if estimator_weights[c] == 0:
# c = c + 1
# continue
# c = c + 1
print("Read model weight {}".format(".\\result\\model_record\\" + model_file))
self.base_estimator.load_weights(".\\result\\model_record\\" + model_file)
result = self.base_estimator.predict(x_test, batch_size=batch_size) / len(file_name)
# result = self.estimator.predict(x_test, batch_size=batch_size) / np.sum(estimator_weights)
print("Result = {}".format(np.sum(result)))
y_predict = y_predict + result
print("Check the threshold.")
y_output = postprocessing.check_threshold(y_predict,
size=self.output_size,
threshold=threshold)
print("Estimate.")
iou = estimate.IOU(y_test, y_output, self.output_size[0], len(y_test))
(precision, recall, F1) = estimate.F1_estimate(y_test, y_output, self.output_size[0], len(y_test))
avr_iou = np.sum(iou) / len(y_test)
avr_precision = np.sum(precision) / len(y_test)
avr_recall = np.sum(recall) / len(y_test)
avr_F1 = np.sum(F1) / len(y_test)
print("Average IOU：{}".format(avr_iou))
print('Save the result.')
mkdir(".\\result\image\\" + self.name)
for index in range(len(y_test)):
img_save = y_output[index] * 255
cv2.imwrite(".\\result\image\\" + self.name + '\\{}.png'.format(data_start + index), img_save)
print('Save image:{}'.format(data_start + index))
ex_iou = excel.Excel()
ex_iou.write_loss_and_iou(save_path, 0, 0, iou, avr_iou)
ex_iou.write_excel("e1", "precision", vertical=True)
ex_iou.write_excel("e2", precision, vertical=True)
ex_iou.write_excel("f1", "avr_precision", vertical=True)
ex_iou.write_excel("f2", avr_precision, vertical=True)
ex_iou.write_excel("g1", "recall", vertical=True)
ex_iou.write_excel("g2", recall, vertical=True)
ex_iou.write_excel("h1", "avr_recall", vertical=True)
ex_iou.write_excel("h2", avr_recall, vertical=True)
ex_iou.write_excel("i1", "F1", vertical=True)
ex_iou.write_excel("i2", F1, vertical=True)
ex_iou.write_excel("j1", "avr_F1", vertical=True)
ex_iou.write_excel("j2", avr_F1, vertical=True)
ex_iou.save_excel(file_name=".\\result\data\\" + save_path + "_iou.xlsx")
ex_iou.close_excel()
def test_weights(self,
x_train,
y_train,
x_test,
y_test,
data_start,
estimator_path,
save_path,
threshold=0.5,
estimator_weights=None):
def mkdir(path):
# 去除首位空格
path = path.strip()
# 去除尾部 \ 符號
path = path.rstrip("\\")
# 判斷路徑是否存在
# 存在 True
# 不存在 False
isExists = os.path.exists(path)
# 判斷結果
if not isExists:
# 如果不存在則建立目錄
print("Building the file.")
# 建立目錄操作函式
os.makedirs(path)
return True
else:
# 如果目錄存在則不建立，並提示目錄已存在
print("File is existing.")
return False
base_estimator = self.base_estimator
# if x_test.ndim == 3:
# x_test = np.expand_dims(x_test, axis=-1)
# if y_test.ndim == 3:
# y_test = np.expand_dims(y_test, axis=-1)
if estimator_weights == None:
estimator_weights = self.update_weight(x_train,
y_train,
estimator_path,
threshold=0.5)
dict_estimators_weights = {}
dict_estimators_weights['estimator_weights'] = estimator_weights
# with open(".\\result\\" + self.name + '_estimator_weights.json', 'w', encoding='utf-8') as f:
# json.dump(dict_estimators_weights, f)
y_predict = np.zeros(y_test.shape, dtype=np.float32)
count = 0
for model_file in estimator_path:
print("Read model weight {}".format(".\\result\\model_record\\" +
model_file + ".h5"))
base_estimator.load_weights(".\\result\\model_record\\" +
model_file + ".h5")
result = base_estimator.predict(
x_test, batch_size=self.batch_size) * estimator_weights[count]
y_predict = y_predict + result
count = count + 1
print("Check the threshold.")
y_output = postprocessing.check_threshold(y_predict,
size=self.output_size,
threshold=threshold)
print("Estimate.")
iou = estimate.IOU(y_test, y_output, self.output_size[0], len(y_test))
(precision, recall, F1) = estimate.F1_estimate(y_test, y_output,
self.output_size[0],
len(y_test))
avr_iou = np.sum(iou) / len(y_test)
avr_precision = np.sum(precision) / len(y_test)
avr_recall = np.sum(recall) / len(y_test)
avr_F1 = np.sum(F1) / len(y_test)
print("Average IOU：{}".format(avr_iou))
print('Save the result.')
mkdir(".\\result\image\\" + save_path)
for index in range(len(y_test)):
img_save = y_output[index] * 255
cv2.imwrite(
".\\result\image\\" + save_path +
'\\{}.png'.format(data_start + index), img_save)
print('Save image:{}'.format(data_start + index))
ex_iou = excel.Excel()
ex_iou.write_loss_and_iou(save_path, 0, 0, iou, avr_iou)
ex_iou.write_excel("e1", "precision", vertical=True)
ex_iou.write_excel("e2", precision, vertical=True)
ex_iou.write_excel("f1", "avr_precision", vertical=True)
ex_iou.write_excel("f2", avr_precision, vertical=True)
ex_iou.write_excel("g1", "recall", vertical=True)
ex_iou.write_excel("g2", recall, vertical=True)
ex_iou.write_excel("h1", "avr_recall", vertical=True)
ex_iou.write_excel("h2", avr_recall, vertical=True)
ex_iou.write_excel("i1", "F1", vertical=True)
ex_iou.write_excel("i2", F1, vertical=True)
ex_iou.write_excel("j1", "avr_F1", vertical=True)
ex_iou.write_excel("j2", avr_F1, vertical=True)
ex_iou.save_excel(file_name=".\\result\data\\" + save_path +
"_AdaBWeightiou.xlsx")
ex_iou.close_excel()
def test_IB(self,
x_train,
y_train,
x_test,
y_test,
data_start,
estimator_path,
save_path,
train_batch,
threshold=0.5):
def mkdir(path):
# 去除首位空格
path = path.strip()
# 去除尾部 \ 符號
path = path.rstrip("\\")
# 判斷路徑是否存在
# 存在 True
# 不存在 False
isExists = os.path.exists(path)
# 判斷結果
if not isExists:
# 如果不存在則建立目錄
print("Building the file.")
# 建立目錄操作函式
os.makedirs(path)
return True
else:
# 如果目錄存在則不建立，並提示目錄已存在
print("File is existing.")
return False
base_estimator = self.base_estimator
# if x_test.ndim == 3:
# x_test = np.expand_dims(x_test, axis=-1)
# if y_test.ndim == 3:
# y_test = np.expand_dims(y_test, axis=-1)
y_predict = np.zeros(y_test.shape, dtype=np.float32)
estimator_weights = np.zeros(len(estimator_path))
# estimator_weights = self.update_weight(x_train, y_train, estimator_path, threshold=0.5)
print("Start IB training processing.")
for index in range(len(x_train) // train_batch):
print("Data for training {}".format(index + 1))
if index > len(x_train) // train_batch:
x = x_train[index * train_batch:]
y = y_train[index * train_batch:]
else:
x = x_train[index * train_batch:(index + 1) * train_batch]
y = y_train[index * train_batch:(index + 1) * train_batch]
print("x {}".format(x.shape))
estomator_weights_update = self.update_weight(x,
y,
estimator_path,
threshold=0.5)
if index == 0:
estimator_weights = estomator_weights_update
else:
estimator_weights = (estimator_weights + index *
estomator_weights_update) / (index + 1)
estimator_weights = estimator_weights / np.sum(estimator_weights)
print("Start predict.")
count = 0
for model_file in estimator_path:
print("Read model weight {}".format(".\\result\\model_record\\" +
model_file + ".h5"))
base_estimator.load_weights(".\\result\\model_record\\" +
model_file + ".h5")
result = base_estimator.predict(
x_test, batch_size=self.batch_size) * estimator_weights[count]
y_predict = y_predict + result
count = count + 1
print("Check the threshold.")
y_output = postprocessing.check_threshold(y_predict,
size=self.output_size,
threshold=threshold)
print("Estimate.")
iou = estimate.IOU(y_test, y_output, self.output_size[0], len(y_test))
(precision, recall, F1) = estimate.F1_estimate(y_test, y_output,
self.output_size[0],
len(y_test))
avr_iou = np.sum(iou) / len(y_test)
avr_precision = np.sum(precision) / len(y_test)
avr_recall = np.sum(recall) / len(y_test)
avr_F1 = np.sum(F1) / len(y_test)
print("Average IOU：{}".format(avr_iou))
print('Save the result.')
mkdir(".\\result\image\\" + save_path)
for index in range(len(y_test)):
img_save = y_output[index] * 255
cv2.imwrite(
".\\result\image\\" + save_path +
'\\{}.png'.format(data_start + index), img_save)
print('Save image:{}'.format(data_start + index))
ex_iou = excel.Excel()
ex_iou.write_loss_and_iou(save_path, 0, 0, iou, avr_iou)
ex_iou.write_excel("e1", "precision", vertical=True)
ex_iou.write_excel("e2", precision, vertical=True)
ex_iou.write_excel("f1", "avr_precision", vertical=True)
ex_iou.write_excel("f2", avr_precision, vertical=True)
ex_iou.write_excel("g1", "recall", vertical=True)
ex_iou.write_excel("g2", recall, vertical=True)
ex_iou.write_excel("h1", "avr_recall", vertical=True)
ex_iou.write_excel("h2", avr_recall, vertical=True)
ex_iou.write_excel("i1", "F1", vertical=True)
ex_iou.write_excel("i2", F1, vertical=True)
ex_iou.write_excel("j1", "avr_F1", vertical=True)
ex_iou.write_excel("j2", avr_F1, vertical=True)
ex_iou.save_excel(file_name=".\\result\data\\" + save_path +
"_iou.xlsx")
ex_iou.close_excel()
def test(self,
x_test,
y_test,
data_start,
estimator_path,
save_path,
batch_size=10,
threshold=0.5):
def mkdir(path):
# 去除首位空格
path = path.strip()
# 去除尾部 \ 符號
path = path.rstrip("\\")
# 判斷路徑是否存在
# 存在 True
# 不存在 False
isExists = os.path.exists(path)
# 判斷結果
if not isExists:
# 如果不存在則建立目錄
print("Building the file.")
# 建立目錄操作函式
os.makedirs(path)
return True
else:
# 如果目錄存在則不建立，並提示目錄已存在
print("File is existing.")
return False
base_estimator = self.base_estimator
# if x_test.ndim == 3:
# x_test = np.expand_dims(x_test, axis=-1)
# if y_test.ndim == 3:
# y_test = np.expand_dims(y_test, axis=-1)
base_estimator.load_weights(".\\result\\model_record\\" +
estimator_path + ".h5")
y_predict = base_estimator.predict(x_test, batch_size=batch_size)
print("Check the threshold.\ny_test.shape = {}".format(y_test.shape))
y_output = postprocessing.check_threshold(y_predict,
size=self.output_size,
threshold=threshold)
print("Estimate.")
iou = estimate.IOU(y_test, y_output, self.output_size[0], len(y_test))
(precision, recall, F1) = estimate.F1_estimate(y_test, y_output,
self.output_size[0],
len(y_test))
avr_iou = np.sum(iou) / len(y_test)
avr_precision = np.sum(precision) / len(y_test)
avr_recall = np.sum(recall) / len(y_test)
avr_F1 = np.sum(F1) / len(y_test)
print("Average IOU：{}".format(avr_iou))
print('Save the result.')
mkdir(".\\result\image\\" + save_path)
for index in range(len(y_test)):
img_save = y_output[index] * 255
cv2.imwrite(
".\\result\image\\" + save_path +
'\\{}.png'.format(data_start + index), img_save)
print('Save image:{}'.format(data_start + index))
ex_iou = excel.Excel()
ex_iou.write_loss_and_iou(save_path, 0, 0, iou, avr_iou)
ex_iou.write_excel("e1", "precision", vertical=True)
ex_iou.write_excel("e2", precision, vertical=True)
ex_iou.write_excel("f1", "avr_precision", vertical=True)
ex_iou.write_excel("f2", avr_precision, vertical=True)
ex_iou.write_excel("g1", "recall", vertical=True)
ex_iou.write_excel("g2", recall, vertical=True)
ex_iou.write_excel("h1", "avr_recall", vertical=True)
ex_iou.write_excel("h2", avr_recall, vertical=True)
ex_iou.write_excel("i1", "F1", vertical=True)
ex_iou.write_excel("i2", F1, vertical=True)
ex_iou.write_excel("j1", "avr_F1", vertical=True)
ex_iou.write_excel("j2", avr_F1, vertical=True)
ex_iou.save_excel(file_name=".\\result\data\\" + save_path +
"_iou.xlsx")
ex_iou.close_excel()