딥러닝 코드 질문

12개월 전

안녕하세요, 현재 딥러닝으로 세포 image segmentaion을 하려고 하는데요. 제가 코드를 돌리면 iou = 0이 뜨고, 아래의 첨부한 그림과 같이 그래프가 뜨질 않습니다. 혹시 어떤 부분을 수정해야할까요?
![](https://devsnote.com/assets/data/ck_upload/_upload_0399ec86cfbb9773701fa679f91b99f4.png?v=1684009377)
![](https://hosoft-bucket001.s3.amazonaws.com/devsnote/_upload_976d203b0fab6605dac8f83dfca6eb65-816.png?v=1684009366)

```
from simple_unet_model import simple_unet_model # Use normal unet model
from keras.utils import normalize
import os
import cv2
from PIL import Image
import numpy as np
from matplotlib import pyplot as plt

image_directory = r"C:\Users\Desktop\VS Code\cell_images"
mask_directory = r"C:\Users\Desktop\VS Code\cell_masks"

SIZE = 256
image_dataset = []
mask_dataset = []

images = os.listdir(image_directory)
for i, image_name in enumerate(images):
    if (image_name.split('.')[1] == 'png'):
        image_path = os.path.join(image_directory, image_name)
        image = cv2.imread(image_path,0)
        if image is None:
            print(f"Unable to read image file: {image_path}")
            continue
        image = Image.fromarray(image)
        image = image.resize((SIZE, SIZE))
        image_dataset.append(np.array(image))
        
masks = os.listdir(mask_directory)
for i, image_name in enumerate(masks):
    if (image_name.split('.')[1] == 'png'):
        image_path = os.path.join(mask_directory, image_name)
        image = cv2.imread(image_path,0)
        if image is None:
            print(f"Unable to read image file: {image_path}")
            continue
        image = Image.fromarray(image)
        image = image.resize((SIZE, SIZE))
        mask_dataset.append(np.array(image))
        
        
# Normalize images
image_dataset = np.expand_dims(normalize(np.array(image_dataset), axis = 1),3)
#D not normalize masks, just rescale to 0 to 1.
mask_dataset = np.expand_dims((np.array(mask_dataset)),3) /255

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(image_dataset, mask_dataset, test_size = 0.10, random_state = 0)

# Sanity check, view few images
import random
import numpy as np
image_number = random.randint(0, len(X_train))
plt.figure(figsize = (12, 6))
plt.subplot(121)
plt.imshow(np.reshape(X_train[image_number], (256, 256)), cmap = 'gray')
plt.subplot(122)
plt.imshow(np.reshape(y_train[image_number], (256, 256)), cmap = 'gray')
plt.show()

#######################################################################
IMG_HEIGHT = image_dataset.shape[1]
IMG_WIDTH = image_dataset.shape[2]
IMG_CHANNELS = image_dataset.shape[3]

def get_model():
    return simple_unet_model(IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS)
model = get_model()

# if starting with pre-trained weights.
# model.Load_weights('mitochondria_gpu_tf1.4.hdf5)

history = model.fit(X_train, y_train,
                    batch_size = 16,
                    verbose = 1,
                    epochs = 1,
                    validation_data = (X_test, y_test),
                    shuffle = False)

model.save('cell_test1.hdf5')

###########################################################
# Evaluate the model

_, accuracy = model.evaluate(X_test, y_test)
print("Accuracy = ", (accuracy * 100.0), "%")

# Plot the training and validation accuracy and Loss at each epoch
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(loss) + 1)
plt.plot(epochs, loss, 'y', label = 'Training loss')
plt.plot(epochs, val_loss, 'r', label = 'Validation loss')
plt.title('Training and validation loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

accuracy = history.history['accuracy']
# acc = accuracy
val_accuracy = history.history['val_accuracy']

plt.plot(epochs, accuracy, 'y', label = 'Training acc')
plt.plot(epochs, val_accuracy, 'r', label = 'Validation acc')
plt.title('Training and validation accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.show()

##################################
#IOU
y_pred=model.predict(X_test)
y_pred_thresholded = y_pred > 0

intersection = np.logical_and(y_test, y_pred_thresholded)
union = np.logical_or(y_test, y_pred_thresholded)
iou_score = np.sum(intersection) / np.sum(union)
print("IoU socre is: ", iou_score)

##############################################################
# predict on a few images

model = get_model()

test_img_number = random.randint(0, len(X_test))
test_img = X_test[test_img_number]
ground_truth = y_test[test_img_number]
test_img_norm = test_img[:,:,0][:,:,None]
test_img_input = np.expand_dims(test_img_norm, 0)
prediction = (model.predict(test_img_input)[0,:,:,0] > 0.2).astype(np.uint8)

test_img_other = cv2.imread(r"C:\Users\Desktop\VS Code\cell_images\37_y.png", 0)
test_img_other_norm = np.expand_dims(normalize(np.array(test_img_other), axis = 1), 2)
test_img_other_norm = test_img_other_norm[:,:,0][:,:,None]
test_img_other_input = np.expand_dims(test_img_other_norm, 0)

prediction_other = (model.predict(test_img_other_input)[0,:,:,0] > 0.2).astype(np.uint8)

plt.figure(figsize = (16, 8))
plt.subplot(231)
plt.title("Testing Image")
plt.imshow(test_img[:,:,0], cmap = 'gray')
plt.subplot(232)
plt.title('Testing Lable')
plt.imshow(ground_truth[:,:,], cmap = 'gray')
plt.subplot(233)
plt.title("Prediction on test image")
plt.imshow(prediction, cmap = 'gray')
plt.subplot(234)
plt.title("External Image")
plt.imshow(test_img_other, cmap = 'gray')
plt.subplot(235)
plt.title('Prediction of external Image')
plt.imshow(prediction_other, cmap = 'gray')
plt.show()
```

딥러닝 u-net

113

뎁스_2599

12개월 전

댓글 3

피토니

epochs를 1로 설정했기 때문에 학습이 제대로 이루어지지 않았을 수 있습니다.

딥러닝 모델을 학습할 때 epochs를 충분히 설정하여 학습을 시켜야 모델이 세포 이미지를 segmenting하는 것을 잘 학습할 수 있습니다. 10이나 20 정도로 올려보시기 바랍니다.

그리고 IoU를 계산하는 부분에서 np.sum() 함수는 전체 배열의 요소를 합산하기 때문에 intersection과 union 배열의 요소들을 모두 합산하고 있습니다.

다음과 같이 각각의 배열에 대해 요소별로 합산해야할 것 같습니다.

intersection = np.sum(intersection)
union = np.sum(union)
iou_score = intersection / union

12개월 전 ∙ 0 ∙ 답글 달기 ∙ 공유 ∙ 신고 ∙ 차단

뎁스_2599

@피토니 안녕하세요, 친절한 설명 감사드립니다. 말씀주신 대로 코드를 변경하여 진행해보았는데, iou 값이 IoU score is: 0.3597073650559449로 혹은 비슷하게 나오는데 이 값을 높이기 위해서는 어떻게 해야하는지 알려주실 수 있을지요?

12개월 전 ∙ 0 ∙ 답글 달기 ∙ 공유 ∙ 신고 ∙ 차단

피토니

@뎁스_2599 https://devsnote.com/post/comment/1110 에 올렸습니다.

12개월 전 ∙ 0 ∙ 답글 달기 ∙ 공유 ∙ 신고 ∙ 차단

뎁스노트 앱으로 보시겠습니까?