early = EarlyStopping(
monitor="val_loss", mode="min", patience=10
) # probably needs to be more patient, but kaggle time is limited
callbacks_list = [checkpoint, early, reduceLROnPlat]
if not os.path.exists(weight_path):
bone_age_model.fit_generator(train_gen,
steps_per_epoch=train_size / 100,
validation_data=(valid_X, valid_Y),
epochs=10,
callbacks=callbacks_list,
verbose=1)
bone_age_model.load_weights(weight_path)
print("Training complete !!!\n")
print("Evaluating model on test data\n")
print("Preparing testing dataset")
test_X, test_Y = next(
datagen(test_df,
path_col='path',
y_col='zscore',
gender_col='male',
batch_size=test_size,
target_size=IMG_SIZE,
color_mode='rgb',
seed=8309))
print("Data prepared")
pred_Y = mu + sigma * bone_age_model.predict(
x=test_X, batch_size=25, verbose=1)
test_Y_months = mu + sigma * test_Y
print("Mean absolute error on test data: " +
str(sk_mae(test_Y_months, pred_Y)))
def main():
os.environ["CUDA_VISIBLE_DEVICES"] = str(0)
start = time.time()
RAND_SEED = 2408
IMG_SIZE = (224, 224)
logger = get_logger()
logger.info("=== Start bone age prediction ===")
# Load metadata from csv
df, mu, sigma = load_data_from_dataframe(logger)
# Split into training testing and validation datasets
train_df, valid_df, test_df = split_dataset(df, logger, seed=RAND_SEED)
train_size = train_df.shape[0]
valid_size = valid_df.shape[0]
test_size = test_df.shape[0]
logger.info("Training images: {}".format(train_size))
logger.info("Validation images: {}".format(valid_size))
logger.info("Testing images: {}".format(test_size))
img_gen = get_img_data_gen()
train_gen = flow_from_dataframe(img_gen,
train_df,
path_col='path',
y_col='zscore',
gender_col='male',
batch_size=32,
seed=RAND_SEED,
img_size=IMG_SIZE)
logger.info("Preparing validation data...")
# Get the validation data
valid_gen = flow_from_dataframe(img_gen,
valid_df,
path_col='path',
y_col='zscore',
gender_col='male',
batch_size=valid_size,
seed=RAND_SEED,
img_size=IMG_SIZE)
valid_X, valid_Y = next(valid_gen)
IMG_SHAPE = valid_X[0][0, :, :, :].shape
logger.info("Image shape: "+str(IMG_SHAPE))
logger.info("Data preproation done")
# Model definition
bone_age_model = get_bone_gender_age_vgg_model(IMG_SHAPE,
logger,
mu,
sigma)
logger.info("=== Star training model ===")
# Model Callbacks
epochs = 20
weight_path = "bone_age_weights_untrainable_VGG16_gender" +\
"_{}_epochs_relu_less_dropout_dense.best.hdf5".format(epochs)
checkpoint = ModelCheckpoint(weight_path,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
save_weights_only=True)
reduceLROnPlat = ReduceLROnPlateau(monitor='val_loss',
factor=0.8,
patience=10,
verbose=1,
mode='auto',
epsilon=0.0001,
cooldown=5,
min_lr=0.0001)
early = EarlyStopping(monitor="val_loss", mode="min", patience=10)
callbacks_list = [checkpoint, early, reduceLROnPlat]
if not os.path.exists(weight_path):
history = \
bone_age_model.fit_generator(train_gen,
steps_per_epoch=train_size/32,
validation_data=(valid_X, valid_Y),
epochs=epochs,
callbacks=callbacks_list,
verbose=1)
with open('history_gender_vgg16_freeze_epoch_{}.p'.format(epochs), 'wb') as f:
pickle.dump(history.history, f)
bone_age_model.load_weights(weight_path)
logger.info("Training complete !!!\n")
# Evaluate model on test dataset
logger.info("Evaluating model on test data ...\n")
logger.info("Preparing testing dataset...")
test_gen = flow_from_dataframe(img_gen,
test_df,
path_col='path',
y_col='zscore',
gender_col='male',
batch_size=test_size,
seed=8309,
img_size=IMG_SIZE)
test_X, test_Y = next(test_gen)
logger.info("Data prepared !!!")
pred_Y = mu+sigma*bone_age_model.predict(x=test_X,
batch_size=25,
verbose=1)
test_Y_months = mu+sigma*test_Y
logger.info("Mean absolute error on test data: "
+ str(sk_mae(test_Y_months, pred_Y)))
fig, ax1 = plt.subplots(1, 1, figsize=(6, 6))
ax1.plot(test_Y_months, pred_Y, 'r.', label='predictions')
ax1.plot(test_Y_months, test_Y_months, 'b-', label='actual')
ax1.legend()
ax1.set_xlabel('Actual Age (Months)')
ax1.set_ylabel('Predicted Age (Months)')
plt.savefig('prediction_gender_epoch_{}.png'.format(epochs))
ord_idx = np.argsort(test_Y)
ord_idx = ord_idx[np.linspace(0, len(ord_idx)-1, num=8).astype(int)]
fig, m_axs = plt.subplots(2, 4, figsize=(16, 32))
for (idx, c_ax) in zip(ord_idx, m_axs.flatten()):
c_ax.imshow(test_X[0][idx, :, :, 0], cmap='bone')
title = 'Age: %2.1f\nPredicted Age: %2.1f\nGender: ' % (test_Y_months[idx], pred_Y[idx])
if test_X[1][idx] == 0:
title += "Female\n"
else:
title += "Male\n"
c_ax.set_title(title)
c_ax.axis('off')
plt.savefig('visulize_xray.png')
# Done
total_sec = time.time() - start
logger.info("Total run took {} (Hours:Min:Sec)".format(str(datetime.timedelta(
seconds=total_sec))))
logger.info("done!")
reduceLROnPlat = ReduceLROnPlateau(monitor='val_loss',
factor=0.8,
patience=10,
verbose=1,
mode='auto',
epsilon=0.0001,
cooldown=5,
min_lr=0.0001)
early = EarlyStopping(monitor="val_loss", mode="min", patience=10)
callbacks_list = [checkpoint, early, reduceLROnPlat]
if not os.path.exists(weight_path):
model.fit_generator(train_gen,
steps_per_epoch=500,
validation_data=(valid_X, valid_Y),
epochs=10,
callbacks=callbacks_list,
verbose=1)
else:
model.load_weights(weight_path)
test_X, test_Y = next(
flow_from_dataframe(test_df,
path_col='path',
y_col='Sign',
target_size=IMG_SIZE,
color_mode='rgb',
batch_size=test_size))
pred_Y = model.predict(test_X, batch_size=25, verbose=1)
print("Mean absolute error on test data: " + str(sk_mae(test_Y, pred_Y)))
