def _test_network(self, fun):
with utils.EnvVars(TF_CPP_MIN_LOG_LEVEL=3):
fun_name = fun.__name__
network_outputs = fun()
checkpoint_path = os.path.join("out", "checkpoint", fun_name,
fun_name + ".ckpt")
feed_dict = get_feed_dict()
checkpoint_path = save_random_checkpoint(network_outputs,
checkpoint_path,
feed_dict)
tf.reset_default_graph()
tf.set_random_seed(0)
command = """
./nnef_tools/convert.py
--input-format tensorflow-py \\
--input-model nnef_tests.activation_export.tf_py_network_test_cases.{fun_name} {cp_path}\\
--output-format nnef \\
--output-model out/nnef/{fun_name}.nnef.tgz \\
--compress \\
--conversion-info""".format(fun_name=fun_name,
cp_path=checkpoint_path)
print(command)
convert_using_command(command)
command = """
./nnef_tools/export_activation.py
--input-format tensorflow-py \\
--input-model nnef_tests.activation_export.tf_py_network_test_cases.{fun_name} {cp_path} \\
--conversion-info out/nnef/{fun_name}.nnef.tgz.conversion.json
""".format(fun_name=fun_name, cp_path=checkpoint_path)
print(command)
export_activation_using_command(command)
def _test_network(self, prototxt_path, model_path):
network = os.path.basename(model_path.rsplit('.', 1)[0])
command = """
./nnef_tools/convert.py --input-format caffe \\
--input-model {prototxt_path} {caffemodel_path} \\
--output-format nnef \\
--output-model out/nnef/{network}.nnef.tgz \\
--compress \\
--conversion-info""".format(
prototxt_path=prototxt_path,
caffemodel_path=model_path,
network=network)
print(command)
convert_using_command(command)
command = """
./nnef_tools/export_activation.py --input-format caffe \\
--input-model {prototxt_path} {caffemodel_path} \\
--conversion-info out/nnef/{network}.nnef.tgz.conversion.json
""".format(prototxt_path=prototxt_path,
caffemodel_path=model_path,
network=network)
print(command)
export_activation_using_command(command)
def _test_model(self, filename, run=True, compare=True, max_val=255.0):
network_name = filename.rsplit('/', 1)[1].rsplit('.', 1)[0].replace(
'.', '_').replace('-', '_')
print(filename)
command = """
./nnef_tools/convert.py --input-format=tensorflow-lite \\
--output-format=nnef \\
--input-model={input} \\
--output-model=out/nnef/{network}.nnef \\
--permissive \\
--io-transform SMART_NHWC_TO_NCHW \\
--conversion-info
""".format(input=filename, network=network_name)
print(command)
convert.convert_using_command(command)
command = """
./nnef_tools/convert.py --input-format=nnef \\
--output-format=tensorflow-lite \\
--input-model=out/nnef/{network}.nnef \\
--output-model=out/tflite/{network}.tflite \\
--permissive \\
--io-transform SMART_NCHW_TO_NHWC \\
--conversion-info
""".format(network=network_name)
print(command)
convert.convert_using_command(command)
activation_testing = int(os.environ.get('NNEF_ACTIVATION_TESTING',
'1'))
print("Activation testing is", "ON" if activation_testing else "OFF")
if activation_testing:
import numpy as np
output, output2, input = None, None, None
if run:
output, input = self.run_model(model_path=filename,
max_val=max_val)
output2, _ = self.run_model(
model_path="out/tflite/{}.tflite".format(network_name),
input_data=input,
max_val=max_val)
if compare:
print('Compare:')
print(output.shape, np.min(output), np.mean(output),
np.max(output))
print(output2.shape, np.min(output2), np.mean(output2),
np.max(output2))
self.assertTrue(np.all(np.isfinite(output)))
self.assertTrue(np.all(np.isfinite(output2)))
self.assertTrue(np.allclose(output, output2, atol=1e-5))
def _test_model(self, filename, run=True, compare=True, max_val=255.0):
output, output2, input = None, None, None
if run:
output, input = self.run_model(model_path=filename,
max_val=max_val)
network_name = filename.rsplit('/', 1)[1].rsplit('.', 1)[0].replace(
'.', '_').replace('-', '_')
print(filename)
command = """
./nnef_tools/convert.py --input-framework=tensorflow-lite \\
--output-framework=nnef \\
--input-model={input} \\
--output-directory=out/nnef/{network} \\
--permissive
""".format(input=filename, network=network_name)
print(command)
convert.convert_using_command(command)
command = """
./nnef_tools/convert.py --input-framework=nnef \\
--output-framework=tensorflow-lite \\
--input-model=out/nnef/{network}/model \\
--output-directory=out/tflite/{network} \\
--permissive
""".format(network=network_name)
print(command)
convert.convert_using_command(command)
if run:
output2, _ = self.run_model(
model_path="out/tflite/{}/model.tflite".format(network_name),
input_data=input,
max_val=max_val)
if compare:
print('Compare:')
print(output.shape, np.min(output), np.mean(output),
np.max(output))
print(output2.shape, np.min(output2), np.mean(output2),
np.max(output2))
self.assertTrue(np.all(np.isfinite(output)))
self.assertTrue(np.all(np.isfinite(output2)))
self.assertTrue(np.allclose(output, output2, atol=1e-5))
def _test_network(self, path, size):
network = os.path.basename(path.rsplit('.', 1)[0])
input_shape = "(float32, [2, {size}, {size}, 3])".format(size=size)
command = """
./nnef_tools/convert.py --input-framework=tensorflow-pb \\
--input-model={} \\
--input-shape="{}" \\
--output-framework=nnef \\
--output-directory=out/nnef/{} \\
--compress""".format(path, input_shape,
network)
print(command)
convert.convert_using_command(command)
command = """
./nnef_tools/convert.py --input-framework=nnef \\
--input-model=out/nnef/{}/model.nnef.tgz \\
--output-framework=tensorflow-pb \\
--output-directory=out/tensorflow-pb/{}""".format(
network, network)
print(command)
convert.convert_using_command(command)
tf.reset_default_graph()
load_graph_from_pb(path)
[input] = get_placeholders()
outputs = get_tensors_with_no_consumers()
feed = np.random.random([2, size, size, 3])
with tf.Session() as sess:
activations = sess.run(outputs, feed_dict={input: feed})
tf.reset_default_graph()
load_graph_from_pb('out/tensorflow-pb/{}/model.pb'.format(network))
[input] = get_placeholders()
outputs = get_tensors_with_no_consumers()
with tf.Session() as sess:
activations2 = sess.run(outputs, feed_dict={input: feed})
for a1, a2 in zip(activations, activations2):
self.assertTrue(np.allclose(a1, a2))
def _test_network(self, path):
network = os.path.basename(path.rsplit('.', 1)[0])
command = """
./nnef_tools/convert.py --input-format=tensorflow-pb \\
--input-model={path} \\
--output-format=nnef \\
--output-model=out/nnef/{network}.nnef.tgz \\
--compress \\
--conversion-info""".format(path=path, network=network)
print(command)
convert_using_command(command)
command = """
./nnef_tools/export_activation.py --input-format tensorflow-pb \\
--input-model {path} \\
--conversion-info out/nnef/{network}.nnef.tgz.conversion.json
""".format(path=path, network=network)
print(command)
export_activation_using_command(command)
def _test_model(self,
filename,
run=True,
compare=True,
source_shape="None",
custom_converters=None):
if custom_converters is None:
custom_converters = []
convs = ["onnx_to_nnef_" + conv for conv in custom_converters]
network_name = filename.rsplit('/', 1)[1].rsplit('.', 1)[0].replace(
'.', '_').replace('-', '_')
print(filename)
command = """
./nnef_tools/convert.py --input-framework=onnx \\
--output-framework=nnef \\
--input-model={} \\
--output-directory=out/nnef/{} \\
--input-shape="{}" \\
--custom-converters="{}" \\
--permissive
""".format(filename, network_name, source_shape, ','.join(convs))
print(command)
convert.convert_using_command(command)
convs = ["nnef_to_onnx_" + conv for conv in custom_converters]
command = """
./nnef_tools/convert.py --input-framework=nnef \\
--output-framework=onnx \\
--input-model=out/nnef/{}/model \\
--output-directory=out/onnx/{} \\
--custom-converters="{}" \\
--permissive
""".format(network_name, network_name, ','.join(convs))
print(command)
convert.convert_using_command(command)
filename2 = os.path.join('out', 'onnx', network_name, 'model.onnx')
check_onnx_model(filename2)
g = onnx_io.read_onnx_from_protobuf(filename2)
input_shapes = [input.shape for input in g.inputs]
input_dtypes = [input.dtype for input in g.inputs]
del g
if run:
inputs = []
for input_shape, input_dtype in zip(input_shapes, input_dtypes):
if input_dtype == 'FLOAT':
inputs.append(
np.random.random(input_shape).astype(np.float32) *
0.8 + 0.1)
elif input_dtype == 'BOOL':
inputs.append(np.random.random(input_shape) > 0.5)
elif input_dtype == 'INT64':
inputs.append((np.random.random(input_shape) *
1000).astype(np.int32))
else:
assert False
outputs = None
if compare:
print('Running original ONNX:')
outputs = run_onnx_model(filename, inputs)
print('Running our ONNX:')
outputs2 = run_onnx_model(filename2, inputs)
if compare:
print('Comparing:')
for output, output2 in zip(outputs, outputs2):
# print('Max dist:', np.max(np.abs(output-output2)))
self.assertTrue(np.all(np.isfinite(output)))
self.assertTrue(np.all(np.isfinite(output2)))
self.assertTrue(np.allclose(output, output2, atol=1e-5))
def _test_model(self,
predict_net_path,
init_net_path,
value_info_path,
feed_dict_override=None,
test_shapes=True,
test_outputs=True,
can_run=True,
can_compare=True,
can_convert=True):
network_name = utils.split_path(predict_net_path)[-2]
print('Testing {}: {}, {}, {}'.format(network_name, predict_net_path,
init_net_path, value_info_path))
reader = Reader()
g = reader(predict_net_path, init_net_path, value_info_path)
input_name_shape_dtypes = [(tensor.name, tensor.shape, tensor.dtype)
for tensor in g.inputs]
output_shapes = [t.shape for t in g.outputs]
our_dir = os.path.join('out', 'caffe2', network_name)
if can_convert:
print("Converting...")
nnef_path = os.path.join('out', 'nnef', network_name + '.nnef')
command = """
./nnef_tools/convert.py --input-format caffe2 \\
--output-format nnef \\
--input-model {predict_net} {init_net} {value_info} \\
--output-model {nnef}
""".format(predict_net=predict_net_path,
init_net=init_net_path,
value_info=value_info_path,
nnef=nnef_path)
print(command)
convert_using_command(command)
command = """
./nnef_tools/convert.py --input-format nnef \\
--output-format caffe2 \\
--input-model {nnef} \\
--output-model {out_dir}
""".format(nnef=nnef_path, out_dir=our_dir)
print(command)
convert_using_command(command)
else:
print("Writing out model without conversion...")
writer = Writer()
writer(g, our_dir)
del g
activation_testing = int(os.environ.get('NNEF_ACTIVATION_TESTING',
'1'))
if not activation_testing:
print("Activation testing is OFF")
if can_run and activation_testing:
from caffe2.python import workspace
feed_dict = self._get_random_feed_dict(input_name_shape_dtypes)
if feed_dict_override:
feed_dict.update(feed_dict_override)
print('Running original Caffe2 model...')
outputs = run_caffe2_model(predict_net_path, init_net_path,
feed_dict)
if can_convert:
print('Running converted Caffe2 model...')
else:
print('Running generated Caffe2 model...')
feed_dict2 = {
k.replace('/', '_'): v
for k, v in six.iteritems(feed_dict)
}
outputs2 = run_caffe2_model(
os.path.join(our_dir, 'predict_net.pb'),
os.path.join(our_dir, 'init_net.pb'), feed_dict2)
if can_compare:
print("Comparing...")
self.assertEqual(
{
k.replace('/', '_'): v
for k, v in six.iteritems(
json_utils.load(value_info_path))
}, json_utils.load(os.path.join(our_dir,
'value_info.json')))
for output, output2, output_shape in zip(
outputs, outputs2, output_shapes):
if test_shapes:
self.assertEqual(tuple(output_shape), output.shape)
self.assertEqual(tuple(output_shape), output2.shape)
if test_outputs:
self.assertTrue(np.all(np.isfinite(output)))
self.assertTrue(np.all(np.isfinite(output2)))
self.assertTrue(np.allclose(output, output2,
atol=1e-5))
print('Passed.\n\n')
def _test(self,
fun,
cmp=True,
custom_tf_to_nnef_converters="",
custom_nnef_to_tf_converters="",
test_module="nnef_tests.conversion.tf_py_layer_test_cases",
atol=1e-5):
activation_testing = int(os.environ.get('NNEF_ACTIVATION_TESTING',
'1'))
print("Activation testing is", "ON" if activation_testing else "OFF")
out_dir = os.path.join("out", fun.__name__)
try:
tf.reset_default_graph()
tf.set_random_seed(0)
network_outputs = fun()
feed_dict = get_feed_dict()
old_names = [
placeholder.name for placeholder in get_placeholders()
]
checkpoint_path = os.path.join("out", fun.__name__,
"orig_checkpoint",
fun.__name__ + ".ckpt")
checkpoint_path = save_random_checkpoint(network_outputs,
checkpoint_path,
feed_dict)
tf.reset_default_graph()
tf.set_random_seed(0)
compress_nnef = False
command = """
./nnef_tools/convert.py --input-format tensorflow-py \\
--output-format nnef \\
--input-model {module}.{network} {checkpoint} \\
--output-model out/{network}/{network}.nnef{tgz} \\
--custom-converters {custom} \\
--permissive \\
--io-transformation SMART_TF_NHWC_TO_NCHW \\
{compress}
""".format(checkpoint=checkpoint_path if checkpoint_path else "",
network=fun.__name__,
custom=" ".join(custom_tf_to_nnef_converters),
compress="--compress" if compress_nnef else "",
module=test_module,
tgz=".tgz" if compress_nnef else "")
convert.convert_using_command(command)
if activation_testing:
tf.reset_default_graph()
tf.set_random_seed(0)
network_outputs = fun()
network_output_list = []
utils.recursive_visit(network_outputs,
lambda t: network_output_list.append(t))
# Flatten is needed because of MaxPoolWithArgMax objects
outputs = utils.flatten(
self._run_tfpy(network_output_list, feed_dict,
checkpoint_path))
else:
outputs = None
prefer_nhwc_options = [True]
if tf_has_cuda_gpu():
prefer_nhwc_options += [False]
for prefer_nhwc in prefer_nhwc_options:
print("Converting to TensorFlow {}".format(
"NHWC" if prefer_nhwc else "NCHW"))
data_format_str = ("nhwc" if prefer_nhwc else "nchw")
tf_output_path = os.path.join(
"out", fun.__name__,
fun.__name__ + '_' + data_format_str + '.py')
command = """
./nnef_tools/convert.py --input-format nnef \\
--output-format tensorflow-py \\
--input-model out/{network}/{network}.nnef{tgz} \\
--output-model {output} \\
--io-transformation SMART_NCHW_TO_TF_NHWC \\
--custom-converters {custom} \\
--permissive
""".format(network=fun.__name__,
custom=" ".join(custom_nnef_to_tf_converters),
tgz=".nnef.tgz" if compress_nnef else "",
output=tf_output_path)
convert.convert_using_command(command)
with open(os.path.join(tf_output_path), 'r') as f:
tf_src = f.read()
# noinspection PyProtectedMember
new_net_fun = tf_py_io._tfsource_to_function(
tf_src, fun.__name__)
tf.reset_default_graph()
tf.set_random_seed(0)
if activation_testing:
tf.reset_default_graph()
tf.set_random_seed(0)
network_outputs = new_net_fun()
network_output_list = []
utils.recursive_visit(
network_outputs,
lambda t: network_output_list.append(t))
feed_dict2 = {
placeholder.name: feed_dict[old_names[i]]
for i, placeholder in enumerate(get_placeholders())
}
outputs2 = utils.flatten(
self._run_tfpy(
network_output_list, feed_dict2,
(os.path.join(tf_output_path + ".checkpoint")
if checkpoint_path else None)))
if cmp:
self.assertTrue(len(outputs) == len(outputs2))
for a, b in zip(outputs, outputs2):
if a.dtype == np.bool:
self.assertTrue(np.all(a == b))
else:
print('Max diff:', np.max(np.abs(a - b)))
self.assertTrue(np.all(np.isfinite(a)))
self.assertTrue(np.all(np.isfinite(b)))
self.assertTrue(np.allclose(a, b, atol=atol))
finally:
if self.delete_dats_and_checkpoints:
dat_files = utils.recursive_glob(out_dir, "*.dat")
checkpoints = utils.recursive_glob(out_dir, "*ckpt*")
for file_name in set(dat_files + checkpoints):
os.remove(file_name)
def _test_model(self, prototxt_path, model_path, nonnegative_input=False):
network_name = utils.split_path(prototxt_path)[-2]
print('Testing {}: {}, {}'.format(network_name, prototxt_path,
model_path))
reader = Reader()
g = reader(prototxt_path, model_path)
writer = Writer()
our_prototxt_path = os.path.join('out', 'caffe-ours', network_name,
network_name + '.prototxt')
our_model_path = os.path.join('out', 'caffe-ours', network_name,
network_name + '.caffemodel')
writer(g, our_prototxt_path)
del g
nnef_path = os.path.join('out', 'nnef', network_name + '.nnef')
command = """
./nnef_tools/convert.py --input-format caffe \\
--output-format nnef \\
--input-model {prototxt} {caffemodel} \\
--output-model {nnef}
""".format(prototxt=prototxt_path,
caffemodel=model_path,
nnef=nnef_path)
print(command)
convert_using_command(command)
converted_prototxt_path = os.path.join('out', 'caffe-converted',
network_name,
network_name + '.prototxt')
converted_model_path = os.path.join('out', 'caffe-converted',
network_name,
network_name + '.caffemodel')
command = """
./nnef_tools/convert.py --input-format nnef \\
--output-format caffe \\
--input-model {nnef} \\
--output-model {prototxt}
""".format(nnef=nnef_path, prototxt=converted_prototxt_path)
print(command)
convert_using_command(command)
if int(os.environ.get('NNEF_ACTIVATION_TESTING', '1')):
with utils.EnvVars(GLOG_minloglevel=3):
import caffe
import numpy as np
def random_inputs(net):
# type: (caffe.Net) -> typing.Dict[str, np.ndarray]
np.random.seed(0)
sub = 0.0 if nonnegative_input else 0.5
return {
input:
np.random.random(list(net.blobs[input].shape)) - sub
for input in net.inputs
}
print("Running original net...")
net = caffe.Net(prototxt_path, model_path, caffe.TEST)
out_orig = net.forward(**random_inputs(net))
print("Running generated net...")
net = caffe.Net(our_prototxt_path, our_model_path, caffe.TEST)
out_ours = net.forward(**random_inputs(net))
self.assertEqual(len(out_orig), len(out_ours))
for orig_name, other_name, orig_arr, other_arr in zip(
out_orig.keys(), out_ours.keys(), out_orig.values(),
out_ours.values()):
self.assertTrue(np.allclose(orig_arr, other_arr))
print("Running converted net...")
net = caffe.Net(converted_prototxt_path, converted_model_path,
caffe.TEST)
out_converted = net.forward(**random_inputs(net))
self.assertEqual(len(out_orig), len(out_converted))
for orig_name, other_name, orig_arr, other_arr in zip(
out_orig.keys(), out_converted.keys(),
out_orig.values(), out_converted.values()):
self.assertTrue(
np.allclose(orig_arr, other_arr, rtol=1e-5, atol=1e-5))
print('Done')
def _test_model(self,
filename,
run=True,
compare=True,
source_shape="None",
custom_converters=None):
if custom_converters is None:
custom_converters = []
convs = ["custom.onnx_to_nnef_" + conv for conv in custom_converters]
network_name = filename.rsplit('/', 1)[1].rsplit('.', 1)[0].replace(
'.', '_').replace('-', '_')
print(filename)
command = """
./nnef_tools/convert.py --input-format onnx \\
--output-format nnef \\
--input-model {} \\
--output-model out/nnef/{}.nnef \\
--input-shape "{}" \\
--custom-converters {} \\
--permissive \\
--conversion-info
""".format(filename, network_name, source_shape, ' '.join(convs))
print(command)
convert.convert_using_command(command)
convs = ["custom.nnef_to_onnx_" + conv for conv in custom_converters]
command = """
./nnef_tools/convert.py --input-format nnef \\
--output-format onnx \\
--input-model out/nnef/{}.nnef \\
--output-model out/onnx/{}.onnx \\
--custom-converters {} \\
--permissive \\
--conversion-info
""".format(network_name, network_name, ' '.join(convs))
print(command)
convert.convert_using_command(command)
filename2 = os.path.join('out', 'onnx', network_name + '.onnx')
g = onnx_io.read_onnx_from_protobuf(filename2)
input_shapes = [input.shape for input in g.inputs]
input_dtypes = [input.dtype for input in g.inputs]
del g
activation_testing = int(os.environ.get('NNEF_ACTIVATION_TESTING',
'1'))
print("Activation testing is", "ON" if activation_testing else "OFF")
if activation_testing:
import numpy as np
import caffe2.python.onnx.backend as backend
import onnx
def check_onnx_model(filename):
model = onnx.load(filename)
onnx.checker.check_model(model)
def run_onnx_model(filename, input):
model = onnx.load(filename)
try:
rep = backend.prepare(model, device="CUDA:0")
outputs = rep.run(input)
except Exception:
print("Couldn't run in CUDA, running on CPU:")
rep = backend.prepare(model, device="CPU")
outputs = rep.run(input)
return outputs
check_onnx_model(filename2)
if run:
inputs = []
for input_shape, input_dtype in zip(input_shapes,
input_dtypes):
if input_dtype == 'FLOAT':
inputs.append(
np.random.random(input_shape).astype(np.float32) *
0.8 + 0.1)
elif input_dtype == 'BOOL':
inputs.append(np.random.random(input_shape) > 0.5)
elif input_dtype == 'INT64':
inputs.append((np.random.random(input_shape) *
1000).astype(np.int64))
else:
assert False
outputs = None
if compare:
print('Running original ONNX:')
outputs = run_onnx_model(filename, inputs)
print('Running converted ONNX model...')
outputs2 = run_onnx_model(filename2, inputs)
if compare:
print('Comparing:')
for output, output2 in zip(outputs, outputs2):
# print('Max dist:', np.max(np.abs(output-output2)))
self.assertTrue(np.all(np.isfinite(output)))
self.assertTrue(np.all(np.isfinite(output2)))
self.assertTrue(np.allclose(output, output2,
atol=1e-5))
def _test_network(self, path, source_shape=1, ignore_extra_outputs=False):
assert utils.is_anyint(source_shape) or isinstance(
source_shape, (list, tuple))
network = os.path.basename(path.rsplit('/', 2)[1])
command = """
./nnef_tools/convert.py --input-format=tensorflow-pb \\
--input-model={path} \\
--output-format=nnef \\
--output-model=out/nnef/{network}.nnef.tgz \\
--compress \\
--input-shape="{shape}" """.format(
path=path, network=network, shape=source_shape)
print(command)
convert.convert_using_command(command)
for prefer_nchw in [False, True]:
print("Prefer NCHW" if prefer_nchw else "Prefer NHWC")
prefer_nchw_opt = "--prefer-nchw" if prefer_nchw else ""
prefer_nchw_str = "_prefer_nchw" if prefer_nchw else ""
command = """
./nnef_tools/convert.py --input-format=nnef \\
--input-model=out/nnef/{network}.nnef.tgz \\
--output-format=tensorflow-pb \\
--output-model=out/tensorflow-pb{nchw_str}/{network}.pb \\
{nchw_opt}""".format(
network=network,
nchw_str=prefer_nchw_str,
nchw_opt=prefer_nchw_opt)
print(command)
convert.convert_using_command(command)
activation_testing = int(
os.environ.get('NNEF_ACTIVATION_TESTING', '1'))
print("Activation testing is",
"ON" if activation_testing else "OFF")
if activation_testing:
import numpy as np
import tensorflow as tf
def normalize_shape(shape, default=1):
return [
int(dim.value) if dim.value is not None else default
for dim in shape.dims
]
tf.reset_default_graph()
self._set_default_graph_from_pb(path)
if isinstance(source_shape, (list, tuple)):
feed_dict = {
placeholder.name: np.random.random(source_shape)
for placeholder in get_placeholders()
}
else:
feed_dict = {
placeholder.name: np.random.random(
normalize_shape(placeholder.shape,
default=source_shape))
for placeholder in get_placeholders()
}
old_names = [
placeholder.name for placeholder in get_placeholders()
]
outputs = self._run_tfpb(path, feed_dict)
tf.reset_default_graph()
path2 = os.path.join('out', 'tensorflow-pb' + prefer_nchw_str,
network + ".pb")
self._set_default_graph_from_pb(path2)
feed_dict2 = {
placeholder.name: feed_dict[old_names[i]]
for i, placeholder in enumerate(get_placeholders())
}
outputs2 = self._run_tfpb(path2, feed_dict2)
if ignore_extra_outputs:
outputs2 = outputs2[-len(outputs):]
self.assertTrue(len(outputs) == len(outputs2))
for a, b in zip(outputs, outputs2):
if a.dtype == np.bool:
self.assertTrue(np.all(a == b))
else:
self.assertTrue(np.all(np.isfinite(a)))
self.assertTrue(np.all(np.isfinite(b)))
self.assertTrue(np.allclose(a, b, atol=1e-5))
def _test(self,
fun,
cmp=True,
custom_tf_to_nnef_converters="",
custom_nnef_to_tf_converters="",
test_module="tests.activation.tf_py_layer_test_cases"):
out_dir = os.path.join("out", fun.__name__)
try:
tf.reset_default_graph()
tf.set_random_seed(0)
network_outputs = fun()
feed_dict = get_feed_dict()
checkpoint_path = os.path.join("out", fun.__name__,
"orig_checkpoint",
fun.__name__ + ".ckpt")
checkpoint_path = save_random_checkpoint(network_outputs,
checkpoint_path,
feed_dict)
tf.reset_default_graph()
tf.set_random_seed(0)
compress_nnef = False
command = """
./nnef_tools/convert.py --input-framework=tensorflow-py \\
--output-framework=nnef \\
--input-model={module}.{network}{checkpoint} \\
--output-directory=out/{network}/nnef \\
--custom-converters="{custom}" \\
--permissive \\
--io-transformation=SMART_TF_NHWC_TO_NCHW \\
{compress}
""".format(checkpoint=':' +
checkpoint_path if checkpoint_path else "",
network=fun.__name__,
custom=custom_tf_to_nnef_converters,
compress="--compress" if compress_nnef else "",
module=test_module)
convert.convert_using_command(command)
open(os.path.join("out", fun.__name__, "__init__.py"), "w").close()
tf.reset_default_graph()
tf.set_random_seed(0)
fun()
conv_info = conversion_info.load(
os.path.join("out", fun.__name__, "nnef", "conversion.json"))
tf_activation_exporter.export(output_path=os.path.join(
"out", fun.__name__, "nnef", "activations"),
feed_dict=feed_dict,
conversion_info=conv_info,
checkpoint_path=checkpoint_path,
verbose=False)
prefer_nhwc_options = [True]
if tf_has_cuda_gpu():
prefer_nhwc_options += [False]
for prefer_nhwc in prefer_nhwc_options:
print("Converting to TensorFlow {}".format(
"NHWC" if prefer_nhwc else "NCHW"))
data_format_str = ("nhwc" if prefer_nhwc else "nchw")
tf_output_dir = os.path.join("out", fun.__name__,
"tf_" + data_format_str)
command = """
./nnef_tools/convert.py --input-framework=nnef \\
--output-framework=tensorflow-py \\
--input-model=out/{network}/nnef/model{tgz} \\
--output-directory={output} \\
--io-transformation=SMART_NCHW_TO_TF_NHWC \\
--custom-converters="{custom}" \\
--permissive
""".format(network=fun.__name__,
custom=custom_nnef_to_tf_converters,
tgz=".nnef.tgz" if compress_nnef else "",
output=tf_output_dir)
convert.convert_using_command(command)
open(os.path.join(tf_output_dir, "__init__.py"), "w").close()
open(os.path.join(tf_output_dir, "model", "__init__.py"),
"w").close()
with open(os.path.join(tf_output_dir, "model",
"model.py")) as f:
tf_src = f.read()
# noinspection PyProtectedMember
new_net_fun = tf_py_io._tfsource_to_function(
tf_src, fun.__name__)
conv_info_tf_to_nnef = conversion_info.load(
os.path.join(out_dir, "nnef", "conversion.json"))
conv_info_nnef_to_tf = conversion_info.load(
os.path.join(tf_output_dir, "conversion.json"))
conv_info_tf_to_tf = conversion_info.compose(
conv_info_tf_to_nnef, conv_info_nnef_to_tf)
conversion_info.dump(
conv_info_tf_to_tf,
os.path.join(tf_output_dir, "conv_info_tf_to_tf.json"))
feed_dict2 = activation_test.transform_feed_dict(
feed_dict, conv_info_tf_to_tf)
nnef2_out_dir = os.path.join(out_dir,
"nnef_from_tf_" + data_format_str)
tf.reset_default_graph()
tf.set_random_seed(0)
command = """
./nnef_tools/convert.py --input-framework=tensorflow-py \\
--output-framework=nnef \\
--input-model={input}{checkpoint} \\
--output-directory={output} \\
--custom-converters="{custom}" \\
--permissive \\
--io-transformation=SMART_TF_NHWC_TO_NCHW \\
{compress}
""".format(checkpoint=(
':' +
(os.path.join(tf_output_dir, "model", "checkpoint",
"model.ckpt") if checkpoint_path else "")),
input=tf_output_dir.replace('/', '.') +
".model.model." + fun.__name__,
custom=custom_tf_to_nnef_converters,
compress="--compress" if compress_nnef else "",
output=nnef2_out_dir)
convert.convert_using_command(command)
conv_info_tf_to_nnef2 = conversion_info.load(
os.path.join(out_dir, "nnef_from_tf_" + data_format_str,
"conversion.json"))
conv_info_nnef_to_nnef = conversion_info.compose(
conv_info_nnef_to_tf, conv_info_tf_to_nnef2)
conversion_info.dump(
conv_info_nnef_to_nnef,
os.path.join(nnef2_out_dir, "conv_info_nnef_to_nnef.json"))
tf.reset_default_graph()
tf.set_random_seed(0)
new_net_fun()
tf_activation_exporter.export(
output_path=os.path.join(nnef2_out_dir, "activations"),
feed_dict=feed_dict2,
conversion_info=conv_info_tf_to_nnef2,
checkpoint_path=(os.path.join(tf_output_dir, "model",
"checkpoint", "model.ckpt")
if checkpoint_path else None),
verbose=False)
if cmp:
activation_test.compare_activation_dirs(
os.path.join(out_dir, "nnef", "activations"),
os.path.join(out_dir,
"nnef_from_tf_" + data_format_str,
"activations"),
conv_info_nnef_to_nnef,
verbose=False)
finally:
if DELETE_DATS_AND_CHECKPOINTS:
dat_files = recursive_glob(out_dir, "*.dat")
checkpoints = recursive_glob(out_dir, "*ckpt*")
for file_name in set(dat_files + checkpoints):
os.remove(file_name)
def _test_network(path, source_shape=1):
assert utils.is_anyint(source_shape) or isinstance(
source_shape, (list, tuple))
network = os.path.basename(path.rsplit('.', 1)[0])
command = """
./nnef_tools/convert.py --input-format=tensorflow-pb \\
--input-model={path} \\
--output-format=nnef \\
--output-model=out/nnef/{network}.nnef.tgz \\
--compress \\
--conversion-info \\
--input-shape="{shape}" """.format(
path=path, network=network, shape=source_shape)
print(command)
convert.convert_using_command(command)
for prefer_nchw in [False, True]:
print("Prefer NCHW" if prefer_nchw else "Prefer NHWC")
prefer_nchw_opt = "--prefer-nchw" if prefer_nchw else ""
prefer_nchw_str = "_prefer_nchw" if prefer_nchw else ""
command = """
./nnef_tools/convert.py --input-format=nnef \\
--input-model=out/nnef/{network}.nnef.tgz \\
--output-format=tensorflow-pb \\
--output-model=out/tensorflow-pb{nchw_str}/{network}.pb \\
--conversion-info \\
{nchw_opt}""".format(
network=network,
nchw_str=prefer_nchw_str,
nchw_opt=prefer_nchw_opt)
print(command)
convert.convert_using_command(command)
command = """
./nnef_tools/convert.py --input-format=tensorflow-pb \\
--input-model=out/tensorflow-pb{nchw_str}/{network}.pb \\
--output-format=nnef \\
--output-model=out/nnef2{nchw_str}/{network}.nnef.tgz \\
--compress \\
--conversion-info""".format(
network=network, nchw_str=prefer_nchw_str)
print(command)
convert.convert_using_command(command)
activation_testing = int(
os.environ.get('NNEF_ACTIVATION_TESTING', '1'))
print("Activation testing is",
"ON" if activation_testing else "OFF")
if activation_testing:
import numpy as np
import tensorflow as tf
from nnef_tools import export_activation
from nnef_tools.activation_export.tensorflow import tf_activation_exporter
def normalize_shape(shape, default=1):
return [
int(dim.value) if dim.value is not None else default
for dim in shape.dims
]
tf.reset_default_graph()
export_activation.tf_set_default_graph_from_pb(path)
if isinstance(source_shape, (list, tuple)):
feed_dict = {
placeholder.name: np.random.random(source_shape)
for placeholder in get_placeholders()
}
else:
feed_dict = {
placeholder.name: np.random.random(
normalize_shape(placeholder.shape,
default=source_shape))
for placeholder in get_placeholders()
}
conv_info_tf_to_nnef = conversion_info.load(
os.path.join("out", 'nnef',
network + ".nnef.tgz.conversion.json"))
tf_activation_exporter.export(
output_path=os.path.join("out", 'nnef', network +
".nnef.tgz.activations"),
feed_dict=feed_dict,
conversion_info=conv_info_tf_to_nnef,
input_output_only=True,
verbose=False)
conv_info_nnef_to_tf = conversion_info.load(
os.path.join('out', 'tensorflow-pb' + prefer_nchw_str,
network + ".pb.conversion.json"))
conv_info_tf_to_tf = conversion_info.compose(
conv_info_tf_to_nnef, conv_info_nnef_to_tf)
feed_dict2 = activation_test.transform_feed_dict(
feed_dict, conv_info_tf_to_tf)
conv_info_tf_to_nnef2 = conversion_info.load(
os.path.join('out', 'nnef2' + prefer_nchw_str,
network + ".nnef.tgz.conversion.json"))
conv_info_nnef_to_nnef = conversion_info.compose(
conv_info_nnef_to_tf, conv_info_tf_to_nnef2)
tf.reset_default_graph()
export_activation.tf_set_default_graph_from_pb(
os.path.join('out', 'tensorflow-pb' + prefer_nchw_str,
network + ".pb"))
tf_activation_exporter.export(
output_path=os.path.join("out", 'nnef2' + prefer_nchw_str,
network +
".nnef.tgz.activations"),
feed_dict=feed_dict2,
conversion_info=conv_info_tf_to_nnef2,
input_output_only=True,
verbose=False)
activation_test.compare_activation_dirs(
os.path.join("out", 'nnef',
network + ".nnef.tgz.activations"),
os.path.join("out", 'nnef2' + prefer_nchw_str,
network + ".nnef.tgz.activations"),
conv_info_nnef_to_nnef,
verbose=False)
def _test(self,
fun,
cmp=True,
custom_tf_to_nnef_converters="",
custom_nnef_to_tf_converters="",
test_module="nnef_tests.conversion.tf_py_layer_test_cases"):
activation_testing = int(os.environ.get('NNEF_ACTIVATION_TESTING',
'1'))
print("Activation testing is", "ON" if activation_testing else "OFF")
out_dir = os.path.join("out", fun.__name__)
try:
tf.reset_default_graph()
tf.set_random_seed(0)
network_outputs = fun()
feed_dict = get_feed_dict()
checkpoint_path = os.path.join("out", fun.__name__,
"orig_checkpoint",
fun.__name__ + ".ckpt")
checkpoint_path = save_random_checkpoint(network_outputs,
checkpoint_path,
feed_dict)
tf.reset_default_graph()
tf.set_random_seed(0)
compress_nnef = False
command = """
./nnef_tools/convert.py --input-format tensorflow-py \\
--output-format nnef \\
--input-model {module}.{network} {checkpoint} \\
--output-model out/{network}/{network}.nnef{tgz} \\
--custom-converters {custom} \\
--permissive \\
--io-transformation SMART_TF_NHWC_TO_NCHW \\
--conversion-info \\
{compress}
""".format(
checkpoint=checkpoint_path if checkpoint_path else "",
network=fun.__name__,
custom=" ".join(custom_tf_to_nnef_converters),
compress="--compress" if compress_nnef else "",
module=test_module,
tgz=".tgz" if compress_nnef else "")
convert.convert_using_command(command)
if activation_testing:
tf.reset_default_graph()
tf.set_random_seed(0)
fun()
conv_info = conversion_info.load(
os.path.join("out", fun.__name__,
fun.__name__ + ".nnef.conversion.json"))
tf_activation_exporter.export(output_path=os.path.join(
"out", fun.__name__, fun.__name__ + ".nnef",
"activations"),
feed_dict=feed_dict,
conversion_info=conv_info,
checkpoint_path=checkpoint_path,
verbose=False)
prefer_nhwc_options = [True]
if tf_has_cuda_gpu():
prefer_nhwc_options += [False]
for prefer_nhwc in prefer_nhwc_options:
print("Converting to TensorFlow {}".format(
"NHWC" if prefer_nhwc else "NCHW"))
data_format_str = ("nhwc" if prefer_nhwc else "nchw")
tf_output_path = os.path.join(
"out", fun.__name__,
fun.__name__ + '_' + data_format_str + '.py')
command = """
./nnef_tools/convert.py --input-format nnef \\
--output-format tensorflow-py \\
--input-model out/{network}/{network}.nnef{tgz} \\
--output-model {output} \\
--io-transformation SMART_NCHW_TO_TF_NHWC \\
--custom-converters {custom} \\
--permissive \\
--conversion-info
""".format(network=fun.__name__,
custom=" ".join(custom_nnef_to_tf_converters),
tgz=".nnef.tgz" if compress_nnef else "",
output=tf_output_path)
convert.convert_using_command(command)
with open(os.path.join(tf_output_path), 'r') as f:
tf_src = f.read()
# noinspection PyProtectedMember
new_net_fun = tf_py_io._tfsource_to_function(
tf_src, fun.__name__)
conv_info_tf_to_nnef = conversion_info.load(
os.path.join("out", fun.__name__,
fun.__name__ + ".nnef.conversion.json"))
conv_info_nnef_to_tf = conversion_info.load(
os.path.join(tf_output_path + ".conversion.json"))
conv_info_tf_to_tf = conversion_info.compose(
conv_info_tf_to_nnef, conv_info_nnef_to_tf)
conversion_info.dump(
conv_info_tf_to_tf,
os.path.join(tf_output_path + ".conv_info_tf_to_tf.json"))
feed_dict2 = activation_test.transform_feed_dict(
feed_dict, conv_info_tf_to_tf)
nnef2_out_dir = os.path.join(tf_output_path + ".nnef")
tf.reset_default_graph()
tf.set_random_seed(0)
command = """
./nnef_tools/convert.py --input-format tensorflow-py \\
--output-format nnef \\
--input-model {input} {checkpoint} \\
--output-model {output} \\
--custom-converters {custom} \\
--permissive \\
--io-transformation SMART_TF_NHWC_TO_NCHW \\
--conversion-info \\
{compress}
""".format(
checkpoint=(os.path.join(tf_output_path + ".checkpoint")
if checkpoint_path else ""),
input=tf_output_path.replace('/', '.')[:-len('.py')] +
"." + fun.__name__,
custom=" ".join(custom_tf_to_nnef_converters),
compress="--compress" if compress_nnef else "",
output=nnef2_out_dir)
convert.convert_using_command(command)
conv_info_tf_to_nnef2 = conversion_info.load(
nnef2_out_dir + ".conversion.json")
conv_info_nnef_to_nnef = conversion_info.compose(
conv_info_nnef_to_tf, conv_info_tf_to_nnef2)
conversion_info.dump(
conv_info_nnef_to_nnef,
os.path.join(nnef2_out_dir +
".conv_info_nnef_to_nnef.json"))
if activation_testing:
tf.reset_default_graph()
tf.set_random_seed(0)
new_net_fun()
tf_activation_exporter.export(
output_path=os.path.join(nnef2_out_dir, "activations"),
feed_dict=feed_dict2,
conversion_info=conv_info_tf_to_nnef2,
checkpoint_path=(os.path.join(tf_output_path +
".checkpoint")
if checkpoint_path else None),
verbose=False)
if cmp:
activation_test.compare_activation_dirs(
os.path.join("out", fun.__name__,
fun.__name__ + ".nnef",
"activations"),
os.path.join(nnef2_out_dir, "activations"),
conv_info_nnef_to_nnef,
verbose=False)
finally:
if self.delete_dats_and_checkpoints:
dat_files = utils.recursive_glob(out_dir, "*.dat")
checkpoints = utils.recursive_glob(out_dir, "*ckpt*")
for file_name in set(dat_files + checkpoints):
os.remove(file_name)