def test_bbox_batches_for_number_0(self):
test_helper.get_test_metadata()
# separated bboxes of 25.png(601): [[60, 11, 24, 50], [87, 9, 24, 50], [113, 7, 21, 50]], size of 1.png: 190 x 75
self._test_bbox_batches(
'number_0',
[87 * 100 / 190, 9 * 100 / 75, 24 * 100 / 190, 50 * 100 / 75],
test_helper.test_data_file_number_0)
def test_batches(self):
data_file_path = test_helper.get_test_metadata()
batch_size, size, B, H, W, C = 2, (416, 416), 5, 13, 13, 10
first_loss_batch = self._calculate_loss_feed_batches(0)
second_loss_batch = self._calculate_loss_feed_batches(1)
third_loss_batch = self._calculate_loss_feed_batches(2)
with self.test_session() as sess:
data_batches, origin_image_shape_batch, image_shape_batch, label_batch, label_bboxes_batch = \
yolo.batches(data_file_path, 5, batch_size, size, num_preprocess_threads=1, channels=3, is_training=False)
loss_feed_batches = yolo.prepare_for_loss(B, batch_size,
label_bboxes_batch,
image_shape_batch,
label_batch)
self.assertEqual(data_batches.get_shape(), (2, 416, 416, 3))
self.assertEqual(loss_feed_batches['probs'].get_shape(),
(2, H * W, B, C))
self.assertEqual(loss_feed_batches['confs'].get_shape(),
(2, H * W, B))
self.assertEqual(loss_feed_batches['coord'].get_shape(),
(2, H * W, B, 4))
self.assertEqual(loss_feed_batches['proid'].get_shape(),
(2, H * W, B, C))
self.assertEqual(loss_feed_batches['areas'].get_shape().as_list(),
[2, H * W, B])
self.assertEqual(loss_feed_batches['upleft'].get_shape(),
(2, H * W, B, 2))
self.assertEqual(loss_feed_batches['botright'].get_shape(),
(2, H * W, B, 2))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
sess.run(tf.local_variables_initializer())
_, lfb = sess.run([data_batches, loss_feed_batches])
all_keys = [
'probs', 'confs', 'coord', 'proid', 'areas', 'upleft',
'botright'
]
for k in all_keys:
print(second_loss_batch)
self.assertAllClose(lfb[k][0], first_loss_batch[k])
self.assertAllClose(lfb[k][1], second_loss_batch[k])
_, lfb = sess.run([data_batches, loss_feed_batches])
for k in all_keys:
self.assertAllClose(lfb[k][0], third_loss_batch[k])
coord.request_stop()
coord.join(threads)
sess.close()
def test_train_length_model(self):
data_file_path = test_helper.get_test_metadata()
config = CNNNSRModelConfig(data_file_path=data_file_path, batch_size=2)
with self.test_session():
model = CNNLengthTrainModel(config)
model.build()
train_op = tf.contrib.layers.optimize_loss(
loss=model.total_loss,
global_step=model.global_step,
learning_rate=0.1,
optimizer=tf.train.MomentumOptimizer(0.5, momentum=0.5))
tf.contrib.slim.learning.train(train_op, None, number_of_steps=2)
def test_bbox_batches(self):
batch_size, size = 2, (28, 28)
with self.test_session() as sess:
data_file_path = test_helper.get_test_metadata()
data_batches, bbox_batches = \
inputs.bbox_batches(data_file_path, batch_size, size, num_preprocess_threads=1, channels=3)
self.assertEqual(data_batches.get_shape(), (2, 28, 28, 3))
self.assertEqual(bbox_batches.get_shape(), (2, 4))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# bbox of 1.png: 246, 77, 173, 223, size of 1.png: 741 x 350
_, bb = sess.run([data_batches, bbox_batches])
self.assertAllClose(bb[0],
[246 / 741, 77 / 350, 173 / 741, 223 / 350])
coord.request_stop()
coord.join(threads)
sess.close()
def test_batches(self):
numbers_labels = lambda numbers: np.concatenate([
one_hot(np.array(numbers) + 1, 11),
np.array([one_hot(11, 11) for _ in range(5 - len(numbers))])
])
max_number_length, expected_length_labels, expected_numbers_labels, expected_numbers_labels_1 = \
5, one_hot(np.array([2, 2]), 5), numbers_labels([1, 9]), numbers_labels([2, 3])
data_file_path = test_helper.get_test_metadata()
batch_size, size = 2, (28, 28)
with self.test_session() as sess:
data_batches, length_label_batches, numbers_label_batches = \
inputs.batches(data_file_path, max_number_length, batch_size, size, num_preprocess_threads=1, channels=3)
self.assertEqual(data_batches.get_shape(), (2, 28, 28, 3))
self.assertEqual(length_label_batches.get_shape(),
(2, max_number_length))
self.assertEqual(numbers_label_batches.get_shape(),
(2, max_number_length, 11))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
batches = []
for i in range(5):
batches.append(
sess.run([
data_batches, length_label_batches,
numbers_label_batches
]))
db, llb, nlb = batches[0]
self.assertAllEqual(llb, expected_length_labels)
self.assertNDArrayNear(nlb[0], expected_numbers_labels, 1e-5)
self.assertNDArrayNear(nlb[1], expected_numbers_labels_1, 1e-5)
coord.request_stop()
coord.join(threads)
sess.close()
def test_evaluation_correct_count(self):
data_file_path = test_helper.get_test_metadata()
config = CNNNSRModelConfig(data_file_path=data_file_path, batch_size=2)
with self.test_session() as sess:
model = CNNNSREvalModel(config)
model.build()
sess.run([
tf.global_variables_initializer(),
tf.local_variables_initializer()
])
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
for i in range(10):
print('batch %s correct count: %s' %
(i, model.correct_count(sess)))
coord.request_stop()
coord.join(threads, stop_grace_period_secs=10)
def _test_bbox_batches(self,
target_bbox,
first_expected_bbox,
data_file_path=None):
data_file_path = test_helper.get_test_metadata(
) if data_file_path is None else data_file_path
batch_size, size = 2, (28, 28)
with self.test_session() as sess:
data_batches, bbox_batches = \
inputs.bbox_batches(data_file_path, batch_size, size, 5,
num_preprocess_threads=1, channels=3, target_bbox=target_bbox)
self.assertEqual(data_batches.get_shape(), (2, 28, 28, 3))
self.assertEqual(bbox_batches.get_shape(), (2, 4))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
_, bb = sess.run([data_batches, bbox_batches])
self.assertAllClose(bb[0], first_expected_bbox)
coord.request_stop()
coord.join(threads)
sess.close()
def create_test_config(self):
data_file_path = test_helper.get_test_metadata()
config = YOLOModelConfig(data_file_path=data_file_path,
net_type='yolo',
batch_size=2)
return config