np.random.seed = seed
# Input dimensions
IMG_WIDTH = 224
IMG_HEIGHT = 224
IMG_CHANNELS = 3
TEST_LENGTH = 2
INPUT_SHAPE = (IMG_WIDTH, IMG_HEIGHT, IMG_CHANNELS)
# Input data
TRAIN_LENGTH = 9000
train_inputs = generate_training_set(TRAIN_LENGTH, IMG_HEIGHT, IMG_WIDTH,
IMG_CHANNELS)
train_labels = generate_labels(TRAIN_LENGTH, IMG_HEIGHT, IMG_WIDTH)
# Image manipulations
train_inputs[:999] = [
noise(noise_type="gauss", image=image) for image in train_inputs[:999]
]
train_inputs[1000:1999] = [
noise(noise_type="s&p", image=image) for image in train_inputs[1000:1999]
]
train_inputs[2000:2999] = [
noise(noise_type="poisson", image=image)
for image in train_inputs[2000:2999]
]
train_inputs[3000:3999] = [
noise(noise_type="speckle", image=image)
for image in train_inputs[3000:3999]
})
model.summary()
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=[dice, 'accuracy', mean_iou])
# Input dimensions
IMG_WIDTH = 256
IMG_HEIGHT = 256
IMG_CHANNELS = 3
ITEM_LENGTH = 10
# Load test data
images = generate_training_set(ITEM_LENGTH, IMG_WIDTH, IMG_HEIGHT,
IMG_CHANNELS)
labels = generate_labels(ITEM_LENGTH, IMG_WIDTH, IMG_HEIGHT)
loss, dice, acc, mean_iou = model.evaluate(images, labels, verbose=2)
print('Restored model, accuracy: {:5.2f}%'.format(100 * acc))
images = np.zeros((1, IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS), dtype=np.uint8)
test_image = plt.imread(
"D:\\Projects\\mgr\\test\\4_test.jpg")[:, :, :IMG_CHANNELS]
test_image = resize(test_image, (IMG_HEIGHT, IMG_WIDTH), preserve_range=True)
images[0] = test_image
preds_test = model.predict(images, verbose=1)
label = labels[0]
pred_label = preds_test[0]
f, axarr = plt.subplots(2, 3)
from config import id2code
from data_generator import generate_training_set, generate_labels, heatmap_to_rgb
import matplotlib.pyplot as plt
from image_preprocessing import mean_filter, gaussian_blur, noise
images = generate_training_set(10, 224, 224, 3)
labels = generate_labels(10, 224, 224)
label = labels[0]
images[:5] = [noise(noise_type="speckle", image=image) for image in images[:5]]
for i in range(10):
f, axarr = plt.subplots(1, 1)
axarr.imshow(images[i])
axarr.set_title("original")
plt.show()