def test_classification_basics(net, shape, weights, outputs, middles):
with tf.Graph().as_default():
inputs = tf.placeholder(tf.float32, [None] + list(shape))
model = net(inputs, is_training=False)
assert isinstance(model, tf.Tensor)
x = np.random.random((1, ) + shape).astype(np.float32) * 255
with tf.Session() as sess:
nets.init(model)
y = model.eval({inputs: model.preprocess(x)})
assert y.shape == (1, 1000)
# Check whether the tensor names match the desired ones
assert 'probs' in model.name # for `model`
assert 'logits' in model.logits.name # for `model.logits`
model_name = model.aliases[0]
for (a, b) in zip(model.get_middles(), middle_names(model_name)[1]):
assert a.name.endswith(b) # for `model.get_middles()`
# Disable the following tests for TF==1.1.0
if LooseVersion(tf.__version__) == LooseVersion('1.1.0'):
return
# Check whether the desired list is returned
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
tf.reset_default_graph()
# Clear GraphDef to avoid `GraphDef cannot be larger than 2GB`
with tf.Graph().as_default():
inputs = tf.placeholder(tf.float32, [None] + list(shape))
# Check whether the desired list is returned under scope functions
with tf.name_scope('a'):
with tf.variable_scope('b'):
with tf.name_scope('c'):
model = net(inputs, is_training=False)
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
with tf.variable_scope('d'):
with tf.name_scope('e'):
with tf.variable_scope('f'):
model = net(inputs, is_training=False)
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
tf.reset_default_graph()
def test_basics(net, shape):
inputs = tf.placeholder(tf.float32, [None] + list(shape))
model = net(inputs, is_training=False)
assert isinstance(model, tf.Tensor)
x = np.random.random((1, ) + shape).astype(np.float32)
with tf.Session() as sess:
nets.init(model)
y = model.eval({inputs: x})
assert y.shape == (1, 1000)
def test_detection_basics(net, shape, stem):
# TODO: Once the roi-pooling dependency is removed,
# FasterRCNN-related tests should be added.
with tf.Graph().as_default():
inputs = tf.placeholder(tf.float32, [None] + list(shape))
model = net(inputs, stem, is_training=False)
assert isinstance(model, tf.Tensor)
x = np.random.random((1, 733, 490, 3)).astype(np.float32) * 255
with tf.Session() as sess:
nets.init(model)
y = model.eval({inputs: model.preprocess(x)})
def test_basics(net, shape):
inputs = tf.placeholder(tf.float32, [None] + list(shape))
model = net(inputs, is_training=False)
assert isinstance(model, tf.Tensor)
x = np.random.random((1,) + shape).astype(np.float32)
with tf.Session() as sess:
nets.init(model)
y = model.eval({inputs: x})
for (a, b) in zip(model.get_middles(), direct(model.aliases[0])[1]):
assert a.name.endswith(b)
assert y.shape == (1, 1000)
def test_detection_basics():
# TODO: Once the roi-pooling dependency is removed,
# FasterRCNN-related tests should be added.
inputs = tf.placeholder(tf.float32, [None, 416, 416, 3])
model1 = nets.YOLOv2(inputs, nets.Darknet19, is_training=False)
assert isinstance(model1, tf.Tensor)
model2 = nets.TinyYOLOv2(inputs, nets.TinyDarknet19, is_training=False)
assert isinstance(model2, tf.Tensor)
x = np.random.random((1, 733, 490, 3)).astype(np.float32) * 255
with tf.Session() as sess:
nets.init([model1, model2])
y1 = model1.eval({inputs: model1.preprocess(x)})
y2 = model2.eval({inputs: model2.preprocess(x)})
def test_classification_basics(net, shape, weights, outputs, middles):
with tf.Graph().as_default():
inputs = tf.placeholder(tf.float32, [None] + list(shape))
model = net(inputs, is_training=False)
assert isinstance(model, tf.Tensor)
# Disable tests whether the desired list is returned for TF==1.1.0
if LooseVersion(tf.__version__) > LooseVersion('1.1.0'):
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
x = np.random.random((1, ) + shape).astype(np.float32) * 255
with tf.Session() as sess:
nets.init(model)
y = model.eval({inputs: model.preprocess(x)})
# The tensor name of `model` should be `probs`.
assert 'probs' in model.name
# The tensor name of `model.logits` should be `logits`.
assert 'logits' in model.logits.name
# The tensor names from `get_middles()` should be the desired ones.
for (a, b) in zip(model.get_middles(), direct(model.aliases[0])[1]):
assert a.name.endswith(b)
# Disable tests whether the desired list is returned for TF==1.1.0
if LooseVersion(tf.__version__) > LooseVersion('1.1.0'):
with tf.name_scope('a'):
with tf.variable_scope('b'):
with tf.name_scope('c'):
model = net(inputs, is_training=False)
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
with tf.variable_scope('d'):
with tf.name_scope('e'):
with tf.variable_scope('f'):
model = net(inputs, is_training=False)
assert len(model.get_weights()) == weights
assert len(model.get_outputs()) == outputs
assert len(model.get_middles()) == middles
assert y.shape == (1, 1000)
