def predict(model: Model, dataset: DatasetBase, restore_path: Path):
restore_path = restore_path.expanduser().absolute()
model.build_graph()
saver = tf.train.Saver(save_relative_paths=True)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver.restore(sess, str(restore_path))
while True:
sess.run(dataset.test_init_op)
try:
pred, x, y = sess.run([model.predict(), dataset.x, dataset.y],
feed_dict={tf.keras.backend.learning_phase(): 0})
pred = np.argmax(pred)
print(pred)
io.imshow(x[0, :, :, 0], cmap="gray")
io.show()
except tf.errors.OutOfRangeError:
continue
parser = argparse.ArgumentParser()
parser.add_argument('image_dir')
parser.add_argument('model')
args = parser.parse_args()
model_path = args.model
data_path = args.image_dir
print('Loading model from ', model_path)
model = Model().load(model_path)
print('Model loaded')
print(model._config)
print('## Evaluating on test data##')
prediction, labels = model.predict(data_path, return_labels=True)
prediction_class = prediction.argmax(axis=-1)
correct = (prediction_class == labels).sum()
total = len(labels)
print('Percentage correct (manual): {:.2f}, {}/{}'.format((correct / total * 100), correct, total))
#np.save('predictions.npy', {'prediction': prediction, 'true': y_test, 'labels': labels})
"""
# Legacy code
print('Loading model from ', model_path)
model = tf.keras.models.load_model(model_path)
