def test_model():
model_name = "fcn_8"
h = 224
w = 256
n_c = 100
check_path = "/tmp/%d" % (random.randint(0, 199999))
m = models.model_from_name[model_name](n_c, input_height=h, input_width=w)
m.train(train_images=tr_im,
train_annotations=tr_an,
steps_per_epoch=2,
epochs=2,
checkpoints_path=check_path)
m.predict_segmentation(np.zeros((h, w, 3))).shape
predict_multiple(inp_dir=te_im,
checkpoints_path=check_path,
out_dir="/tmp")
predict_multiple(inps=[np.zeros((h, w, 3))] * 3,
checkpoints_path=check_path,
out_dir="/tmp")
o = predict(inp=np.zeros((h, w, 3)), checkpoints_path=check_path)
o.shape
def test_model():
model_name = "fcn_8"
h = 224
w = 256
n_c = 100
check_path = tempfile.mktemp()
m = all_models.model_from_name[model_name](n_c, input_height=h, input_width=w)
m.train(train_images=tr_im,
train_annotations=tr_an,
steps_per_epoch=2,
epochs=2,
checkpoints_path=check_path
)
m.predict_segmentation(np.zeros((h, w, 3))).shape
predict_multiple(
inp_dir=te_im, checkpoints_path=check_path, out_dir="/tmp")
predict_multiple(inps=[np.zeros((h, w, 3))]*3,
checkpoints_path=check_path, out_dir="/tmp")
ev = m.evaluate_segmentation( inp_images_dir=te_im , annotations_dir=te_an )
assert ev['frequency_weighted_IU'] > 0.01
print(ev)
o = predict(inp=np.zeros((h, w, 3)), checkpoints_path=check_path)
o.shape
ev = evaluate( inp_images_dir=te_im , annotations_dir=te_an , checkpoints_path=check_path)
assert ev['frequency_weighted_IU'] > 0.01
verify_dataset=True,
# load_weights="weights/vgg_unet_1.4" ,
optimizer_name='adadelta',
do_augment=True,
augmentation_name="aug_all",
checkpoints_path="weights/vgg_unet_1",
epochs=10)
# Display the model's architecture
model.summary()
# Save the entire model to a HDF5 file.
# The '.h5' extension indicates that the model should be saved to HDF5.
model.save('vgg_unet_1.h5')
#predict an image from the training data
out = model.predict_segmentation(checkpoints_path="weights/vgg_unet_1",
inp="dataset1/images_prepped_test/43.jpg",
out_fname="newout.png")
from keras_segmentation.predict import predict_multiple
predict_multiple(checkpoints_path="weights/vgg_unet_1",
inp_dir="dataset1/images_prepped_test/",
out_dir="weights/out/")
#import matplotlib.pyplot as plt
#plt.imshow(out)
#
## evaluating the model
#print(model.evaluate_segmentation( inp_images_dir="dataset1/images_prepped_test/" , annotations_dir="dataset1/annotations_prepped_test/" ) )
def evaluate_model(image_dir,
label_dir=None,
checkpoints_path=None,
calculate_predicting_indicators=True,
output_predicted_result=False,
segment_out_predicted_region_from_original_images=False,
roi_description='roi',
preds=None,
batch_process_slice_point=None,
target_roi_value=None):
from keras_segmentation.predict import predict_multiple
if preds is None:
print('----------生成模型預測結果----------')
preds = predict_multiple(checkpoints_path=checkpoints_path,
inp_dir=image_dir)
preds_batches = []
if batch_process_slice_point:
for i in range(len(batch_process_slice_point) + 1):
if i == 0:
preds_batches.append(preds[:batch_process_slice_point[i]])
elif i == len(batch_process_slice_point):
preds_batches.append(preds[batch_process_slice_point[i - 1]:])
else:
preds_batches.append(preds[
batch_process_slice_point[i -
1]:batch_process_slice_point[i]])
else:
preds_batches.append(preds)
base_index = 0
dice_score_list = []
recall_list = []
precision_list = []
globalDice_part = []
TP = 0
FP = 0
FN = 0
TN = 0
counter1 = 0
counter2 = 0
counter3 = 0
for idx, preds_batch in enumerate(preds_batches):
print(f'----------預測結果資料型態轉換(第{idx + 1}批資料)----------')
preds_batch = np.asarray(preds_batch).astype(np.uint8)
if calculate_predicting_indicators:
labels_dicePerCase = []
preds_dicePerCase = []
labels_globalDice = []
preds_globalDice = []
# 取得該批資料第一位病患的編號
# 如果影像的命名方法為「病患編號_影像編號」，這裡的病患編號為「病患編號」
# 如果影像的命名方法為「資料集_病患編號_影像編號」，這裡的病患編號為「資料集_病患編號」
separator = '_'
prev_patient_idx = separator.join(
os.listdir(label_dir)[base_index + 0].split('_')[:-1])
print(f'----------開始計算各項預測指標(第{idx + 1}批資料)----------')
for i in range(preds_batch.shape[0]):
label = cv2.imread(
os.path.join(label_dir,
os.listdir(label_dir)[base_index + i]),
cv2.IMREAD_GRAYSCALE)
labels_globalDice.append(label)
if preds_batch[i].shape != label.shape:
pred = cv2.resize(
preds_batch[i].copy(),
(label.shape[0],
label.shape[1])) # 預測的標記照片大小必須和原始的標記照片一樣
else:
pred = preds_batch[i].copy()
preds_globalDice.append(pred)
# 計算混淆矩陣
if (1 in np.unique(label)) and (1 in np.unique(pred)):
TP += 1
elif (1 not in np.unique(label)) and (1 in np.unique(pred)):
FP += 1
elif (1 in np.unique(label)) and (1 not in np.unique(pred)):
FN += 1
else:
TN += 1
# 取得目前處理的影像對應的病患編號
separator = '_'
patient_idx = separator.join(
os.listdir(label_dir)[base_index + i].split('_')[:-1])
if patient_idx != prev_patient_idx: # 判斷是否到達下一位病患的影像
labels_dicePerCase = np.asarray(labels_dicePerCase)
preds_dicePerCase = np.asarray(preds_dicePerCase)
# print(f'編號為{patient_idx}病患的CT影像張數：{labels_dicePerCase.shape[0]}') #####
dice_score_list.append(
dice_score(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
recall_list.append(
recall(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
precision_list.append(
precision(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
labels_dicePerCase = []
preds_dicePerCase = []
labels_dicePerCase.append(label)
preds_dicePerCase.append(pred)
if i == preds_batch.shape[
0] - 1: # 判斷是否為最後一張的影像，如果是則開始計算最後一位病患的average Dice score per case
labels_dicePerCase = np.asarray(labels_dicePerCase)
preds_dicePerCase = np.asarray(preds_dicePerCase)
# print(f'編號為{patient_idx}病患的CT影像張數：{labels_dicePerCase.shape[0]}') #####
dice_score_list.append(
dice_score(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
recall_list.append(
recall(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
precision_list.append(
precision(labels_dicePerCase, preds_dicePerCase,
target_roi_value))
prev_patient_idx = patient_idx
counter1 += 1
if counter1 % 500 == 0:
print('目前進度：第' + str(counter1) + '張照片')
### 計算 global Dice (part) ###
labels_globalDice = np.asarray(labels_globalDice)
preds_globalDice = np.asarray(preds_globalDice)
globalDice_part.append(
dice_score(labels_globalDice, preds_globalDice,
target_roi_value))
if output_predicted_result:
if label_dir:
save_path = label_dir + '_predicted'
else:
save_path = image_dir.replace(
image_dir.split('\\')[-1],
'annotations') + '_' + roi_description + '_predicted'
if not os.path.exists(save_path):
os.makedirs(save_path)
print('-----建立新資料夾：' + save_path + '-----')
print(f'---------開始輸出模型預測結果(第{idx + 1}批資料)----------')
for i in range(preds_batch.shape[0]):
image_ori = cv2.imread(
os.path.join(image_dir,
os.listdir(image_dir)[base_index + i]),
cv2.IMREAD_GRAYSCALE)
image_ori_name = os.listdir(image_dir)[base_index + i]
if preds_batch[i].shape != image_ori.shape:
pred = cv2.resize(preds_batch[i],
(image_ori.shape[0], image_ori.shape[1]))
else:
pred = preds_batch[i]
cv2.imwrite(os.path.join(save_path, image_ori_name), pred)
counter2 += 1
if counter2 % 500 == 0:
print('目前進度：第' + str(counter2) + '張照片')
if segment_out_predicted_region_from_original_images:
save_path = image_dir + '_only_containing_predicted_roi_' + roi_description
if not os.path.exists(save_path):
os.makedirs(save_path)
print('-----建立新資料夾：' + save_path + '-----')
print(f'----------開始生成並輸出只包含模型預測區域的圖片(第{idx + 1}批資料)----------')
for i in range(preds_batch.shape[0]):
image_ori = cv2.imread(
os.path.join(image_dir,
os.listdir(image_dir)[base_index + i]),
cv2.IMREAD_GRAYSCALE)
if preds_batch[i].shape != image_ori.shape:
pred = cv2.resize(preds_batch[i],
(image_ori.shape[0], image_ori.shape[1]))
else:
pred = preds_batch[i]
image_pred_roi_region = pred * image_ori # 只保留原始圖片中模型預測區域的位置(其他區域視為像素等於0的背景)
cv2.imwrite(
os.path.join(save_path,
os.listdir(image_dir)[base_index + i]),
image_pred_roi_region)
counter3 += 1
if counter3 % 500 == 0:
print('目前進度：第' + str(counter3) + '張照片')
base_index += len(preds_batch)
### 計算 global Dice ###
if calculate_predicting_indicators:
globalDice = np.mean(globalDice_part)
print(f'total case number: {len(preds)}')
if calculate_predicting_indicators:
return np.mean(dice_score_list), np.mean(recall_list), np.mean(
precision_list
), globalDice, preds, dice_score_list, recall_list, precision_list, TP, FP, FN, TN
else:
return preds
import tensorflow as tf
from keras_segmentation.predict import predict_multiple
from keras_segmentation.predict import model_from_checkpoint_path
gpu_devices = tf.config.experimental.list_physical_devices('GPU')
for device in gpu_devices: tf.config.experimental.set_memory_growth(device, True)
print(f"[INFO] Predicting on test set mask..")
pdr=predict_multiple(
checkpoints_path=r"C:\Users\matte\PycharmProjects\ecographic_breast_nn\segnet\checkpoints\psp_unet",
inp_dir=r"D:\FISICA MEDICA\radiomics_eco\Dataset_BUSI_with_GT\segnet\images_val",
out_dir=r"C:\Users\matte\PycharmProjects\ecographic_breast_nn\segnet\outputs"
)
model=model_from_checkpoint_path(r"C:\Users\matte\PycharmProjects\ecographic_breast_nn\segnet\checkpoints\psp_unet")
print(f"[INFO] Evaluating model..")
print(model.evaluate_segmentation( inp_images_dir=r"D:\FISICA MEDICA\radiomics_eco\Dataset_BUSI_with_GT\segnet\images_val" , annotations_dir=r"D:\FISICA MEDICA\radiomics_eco\Dataset_BUSI_with_GT\segnet\masks_val" ) )
from keras_segmentation.predict import predict_multiple, evaluate, model_from_checkpoint_path
from keras_segmentation.data_utils.visualize_dataset import visualize_segmentation_dataset
import cv2
checkpoints_path = []
checkpoints_path.append("./saveModel/vgg_unet_1")
checkpoints_path.append("./saveModel/vgg_pspnet_1")
checkpoints_path.append("./saveModel/vgg_segnet_1")
checkpoints_path.append("./saveModel/fcn_32_vgg_1")
for ia in checkpoints_path:
predict_multiple(
checkpoints_path=ia,
inp_dir="./demo/test",
out_dir="./demo/predict",
show_legends = True,
class_names = ["sky", "building", "pole", "road", "sidewalk", "vegetation", "traffic_light", "fence", "car", "person", "rider", "static"],
# overlay_img = True, #Chồng 2 ảnh lên nhau
# prediction_width = 1024,
# prediction_height = 768
)
# view summary
# for ia in checkpoints_path:
# model = model_from_checkpoint_path(ia)
# print(model.summary())
# visuallize
# visualize_segmentation_dataset(images_path="./test/example_dataset/images_prepped_train",segs_path="./test/example_dataset/annotations_prepped_train",n_classes=6,no_show=False)
# visualize_segmentation_dataset(images_path="demo/images_preped_test_small/",segs_path="demo/annotations_preped_test_small/",n_classes=6,no_show=True)
# for seg_img, seg_path in visualize_segmentation_dataset(images_path="demo/images_preped_test_small/",segs_path="demo/annotations_preped_test_small/",n_classes=6,no_show=True):
# cv2.imwrite(f"demo/ground_truth/{seg_path.split('/')[-1]}", seg_img)
