def test_avg_pool():
save_dir = "../tfProfiling/avg_pool"
image_input = ImageInput(seed=713,
batch_size=16,
image_h=224,
image_w=224,
image_c=64)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
dummy_var = tf.Variable(tf.zeros(
[2])) # saver barfs if there are no variables in the graph??
avg_pool_output = tf.layers.average_pooling2d(inputs=in_node,
pool_size=[2, 2],
strides=2,
padding="same")
out_node = tf.identity(avg_pool_output, name="output")
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir=save_dir, verbose=False)
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph(skip_intermediate=True)
print(predictions_after_freeze[0].shape)
def test_pool_1():
# [batch, in_height, in_width, in_channels].
image_input = ImageInput(seed=713,
batch_size=1,
image_h=4,
image_w=4,
image_c=2)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
dummy_var = tf.Variable(tf.random_uniform(
[3, 2])) # saver barfs without a variable
constr = NNImageOps(in_node)
constr.set_filter_hw_inout(2, 2, 2, 2)
constr.set_kernel_hw(2, 2)
constr.set_stride_hw(1, 1)
in1 = constr.execute("max_pool")
out_node = tf.identity(in1, name="output") # calc required dims by hand
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir="pool_1")
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph()
print(predictions_after_freeze[0].shape)
def test_conv_2():
image_input = ImageInput(seed=713,
batch_size=4,
image_h=64,
image_w=64,
image_c=4)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
# in_channels must match between input and filter.
# filter shape is [filter_height, filter_width, in_channels, out_channels]
filter_one = tf.Variable(tf.random_uniform([4, 5, image_input.image_c, 2]),
name="filter1")
atrous_one = tf.nn.atrous_conv2d(in_node,
filters=filter_one,
rate=8,
padding='SAME',
name="atrous_one")
filter_two = tf.Variable(tf.random_uniform([31, 31, 2, 1]), name="filter2")
atrous_two = tf.nn.atrous_conv2d(atrous_one,
filters=filter_two,
rate=2,
padding='VALID')
out_node = tf.identity(atrous_two, name="output")
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir="conv_2")
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph()
print(predictions_after_freeze[0].shape)
def test_conv_4():
image_input = ImageInput(seed=713,
batch_size=3,
image_h=4,
image_w=4,
image_c=5)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
constr = NNImageOps(in_node)
constr.set_filter_size([7, 7, 16, 5])
constr.set_output_shape([3, 8, 8, 16])
constr.set_stride_size([1, 2, 2, 1])
in1 = constr.execute("conv2d_transpose") # size is (3, 8, 8, 16)
constr.set_image(in1)
constr.set_filter_size([2, 2, 32, 16])
constr.set_output_shape([3, 8, 8, 32])
constr.set_rate(2)
in02 = constr.execute("atrous_conv2d_transpose") # size is (3,8,8,32)
constr.set_image(in02)
constr.set_filter_size([2, 2, 32, 2])
constr.set_pointwise_filter([1, 1, 64, 8])
constr.set_stride_size([1, 1, 1, 1])
constr.set_rate([1, 2])
in2 = constr.execute("separable_conv2d") # (3, 8, 8, 8)
constr.set_image(in2)
constr.set_filter_size([2, 2, 8, 4])
constr.set_stride_size([1, 1, 1, 1])
constr.set_rate([3, 2])
in3 = constr.execute("depthwise_conv2d") # size (3,8,8,32)
in33 = tf.space_to_batch_nd(in3, [3, 3],
paddings=[[1, 0], [0, 1]]) # size (27,3,3,32)
constr.set_image(tf.reshape(in33, [27 * 2, 12, 12]))
constr.set_filter_size([2, 12, 2])
constr.set_stride_size(1)
in4 = constr.execute("conv1d") # size (54,12,2)
out_node = tf.identity(in4, name="output")
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir="conv_4")
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph()
print(predictions_after_freeze[0].shape)
if __name__ == "main":
test_conv_4()
def test_conv_0():
image_input = ImageInput(seed=713,
batch_size=4,
image_h=28,
image_w=28,
image_c=3)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
constr = NNImageOps(in_node)
constr.set_filter_hw_inout(h=5, w=5, in_ch=3, out_ch=3)
constr.set_kernel_hw(3, 3)
constr.set_stride_hw(3, 3)
in1 = constr.execute("conv2d")
constr.set_image(in1)
in2 = constr.execute("avg_pool")
constr.set_image(in2)
in3 = constr.execute("conv2d")
constr.set_image(in3)
in4 = constr.execute("max_pool")
in5 = constr.flatten_convolution(in4)
out_node = tf.matmul(in5,
tf.Variable(tf.random_uniform([3, 2])),
name="output") # calc required dims by hand
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir="conv_0")
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph()
print(predictions_after_freeze[0].shape)
def test_conv_3():
# [batch, in_height, in_width, in_channels].
image_input = ImageInput(seed=713, batch_size=4, image_h=128, image_w=128, image_c=4)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
# [filter_height, filter_width, in_channels, out_channels]. in_channels must match between input and filter.
filter_one = tf.Variable(tf.random_uniform([4, 5, image_input.image_c]), name="filter1")
dilation_one = tf.nn.dilation2d(in_node, filter=filter_one, strides=[1, 2, 3, 1], rates=[1, 5, 7, 1],
padding='SAME',
name="dilation_one")
filter_two = tf.Variable(tf.random_uniform([11, 7, 4]), name="filter2")
dilation_two = tf.nn.dilation2d(dilation_one, filter=filter_two, strides=[1, 3, 2, 1], rates=[1, 2, 3, 1],
padding='VALID',
name="output")
out_node = tf.identity(dilation_two, name="output")
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir="conv_3")
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph()
print(predictions_after_freeze[0].shape)
def test_conv2d():
save_dir = "../tfProfiling/conv2d"
image_input = ImageInput(seed=713,
batch_size=16,
image_h=256,
image_w=256,
image_c=3)
in_node = image_input.get_placeholder("image", data_type=tf.float32)
conv_2d_output = tf.layers.conv2d(inputs=in_node,
filters=64,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu)
out_node = tf.identity(conv_2d_output, name="output")
placeholders = [in_node]
predictions = [out_node]
tfp = TensorFlowPersistor(save_dir=save_dir, verbose=False)
predictions_after_freeze = tfp \
.set_placeholders(placeholders) \
.set_output_tensors(predictions) \
.set_test_data(image_input.get_test_data()) \
.build_save_frozen_graph(skip_intermediate=True)
print(predictions_after_freeze[0].shape)