from segmentation_models import get_preprocessing
from segmentation_models.losses import bce_jaccard_loss
from segmentation_models.metrics import iou_score
from tensorflow.keras.datasets import mnist
BACKBONE = 'resnet34'
preprocess_input = get_preprocessing(BACKBONE)
# load your data
(x_train, y_train),( x_val, y_val) = mnist.load_data()
# preprocess input
x_train = preprocess_input(x_train)
x_val = preprocess_input(x_val)
# define model
model = FPN(BACKBONE, input_shape=(224, 224, 6),classes=9, encoder_weights=None)
model.compile('Adam', loss=bce_jaccard_loss, metrics=[iou_score])
# fit model
model.fit(
x=x_train,
y=y_train,
batch_size=16,
epochs=100,
validation_data=(x_val, y_val),
)
pyramid_dropout=None,
)
image = np.load("./data/image.npy")
label = np.load("./data/label.npy")
model.compile(
loss=lambda labels, predictions: tf.keras.losses.binary_crossentropy(
labels[:, :, :, 1:], predictions
),
optimizer="adam",
metrics=["accuracy"],
)
model.fit(
[image],
[label],
batch_size=1,
epochs=300,
verbose=2,
callbacks=[
tf.keras.callbacks.ModelCheckpoint(
filepath="./dist/keras/psenet", save_weights_only=True
)
],
)
kernels = model.predict(image)
plt.figure()
plt.imshow(np.squeeze(kernels[:, :, :, 0]))
plt.savefig("./docs/training-with-pure-keras.png")