# read dataset
dataset = Dataset("training-1.csv")
# create column features for further fit them into a model
dataset.create_feature_columns()
# preprocess it
dataset.preprocess()
#split on train/test
dataset.split_data()
# initialize model
model = MyModel()
# build it
model.build(dataset.feature_columns)
# train
model.train(dataset.train_ds, dataset.val_ds)
# evaluate
model.evaluate(dataset.test_ds)
# lr ExponentialDecay
lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
config['learning_rate'],
decay_steps=config['decay_steps'],
decay_rate=config['decay_rate'])
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=lr_schedule),
loss=Loss(),
metrics=[PSNR(), SSIM()],
)
# resume checkpoint
if config['resume']:
model.build((None, None, None, 3))
model.load_weights(checkpoint_path)
# save the best model checkpoint
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_path,
save_weights_only=True,
monitor='val_loss',
mode='min',
save_best_only=True)
# tensorboard visulization
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir,
histogram_freq=1)
# model fit
# This are **roughly** the KITTI camera parameters. But they are not precise, and
# they are not even the same on all sequences!!
fx = 720.0
fy = 720.0
cx = 608.0
cy = 180.0
# I've downsampled the images, so:
fx /= 1.725
fy /= 1.67
cx /= 1.725
cy /= 1.67
model = MyModel(fx, fy, cx, cy)
model.build((None, 224, 720, 6))
model.summary()
if restore_path:
model.load_weights(restore_path)
train_loss = tf.keras.metrics.Mean(name='train_loss')
optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)
# optimizer = tf.keras.optimizers.Adam(learning_rate=1e-4)
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
log_dir = 'logs/' + current_time
summary_writer = tf.summary.create_file_writer(log_dir)
# save_path = 'checkpoints\\run_' + current_time
