def convert_pytorch2keras2ir():
max_error = 0
model = VGG16(vgg([64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'C', 512, 512, 512, 'M', 512, 512, 512], 3))
# load weights here
state_dict_to_load = torch.load('vgg16_reducedfc/vgg16_reducedfc.pth', map_location=lambda storage, loc: storage)
renamed_tate_dict_to_load = {}
for k, v in state_dict_to_load.items():
renamed_tate_dict_to_load['vgg.' + k] = v
print(model.state_dict().keys())
print(state_dict_to_load.keys())
model.load_state_dict(renamed_tate_dict_to_load)
for m in model.modules():
m.training = False
input_np = np.ones((1, 3, 224, 224)) * 0.5
# np.random.uniform(0, 1, (1, 3, 224, 224))
input_var = Variable(torch.FloatTensor(input_np))
output = model(input_var)
k_model = pytorch_to_keras((3, 224, 224,), output)
pytorch_output = output.data.numpy()
print(pytorch_output)
print(np.argmax(pytorch_output))
return
keras_output = k_model.predict(input_np)
error = np.max(pytorch_output - keras_output)
print(error)
if max_error < error:
max_error = error
print('Max error: {0}'.format(max_error))
# save network structure as JSON
json_string = k_model.to_json()
with open("vgg16_reducedfc/imagenet_vgg16_reducedfc.json", "w") as of:
of.write(json_string)
print("Network structure is saved as [vgg16_reducedfc/imagenet_vgg16_reducedfc.json].")
k_model.save_weights('vgg16_reducedfc/imagenet_vgg16_reducedfc.h5')
print("Network weights are saved as [vgg16_reducedfc/imagenet_vgg16_reducedfc.h5].")
parser = Keras2Parser(('vgg16_reducedfc/imagenet_vgg16_reducedfc.json', 'vgg16_reducedfc/imagenet_vgg16_reducedfc.h5'))
parser.run('vgg16_reducedfc/ir')
def KerasParse(architecture_name, image_path):
# get original model prediction result
original_predict = keras_extractor.inference(architecture_name, image_path)
# download model
model_filename = keras_extractor.download(architecture_name, TestModels.cachedir)
# original to IR
parser = Keras2Parser(model_filename)
parser.gen_IR()
parser.save_to_proto(TestModels.tmpdir + architecture_name + "_converted.pb")
parser.save_weights(TestModels.tmpdir + architecture_name + "_converted.npy")
del parser
return original_predict
def KerasParse(architecture_name, image_path):
from mmdnn.conversion.examples.keras.extractor import keras_extractor
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
# get original model prediction result
original_predict = keras_extractor.inference(architecture_name, TestModels.cachedir, image_path)
# download model
model_filename = keras_extractor.download(architecture_name, TestModels.cachedir)
del keras_extractor
# original to IR
IR_file = TestModels.tmpdir + 'keras_' + architecture_name + "_converted"
parser = Keras2Parser(model_filename)
parser.run(IR_file)
del parser
del Keras2Parser
return original_predict
def _convert(args):
if args.srcFramework == 'caffe':
from mmdnn.conversion.caffe.transformer import CaffeTransformer
transformer = CaffeTransformer(args.network,
args.weights,
"tensorflow",
args.inputShape,
phase=args.caffePhase)
graph = transformer.transform_graph()
data = transformer.transform_data()
from mmdnn.conversion.caffe.writer import JsonFormatter, ModelSaver, PyWriter
JsonFormatter(graph).dump(args.dstPath + ".json")
print("IR network structure is saved as [{}.json].".format(
args.dstPath))
prototxt = graph.as_graph_def().SerializeToString()
with open(args.dstPath + ".pb", 'wb') as of:
of.write(prototxt)
print("IR network structure is saved as [{}.pb].".format(args.dstPath))
import numpy as np
with open(args.dstPath + ".npy", 'wb') as of:
np.save(of, data)
print("IR weights are saved as [{}.npy].".format(args.dstPath))
return 0
elif args.srcFramework == 'caffe2':
raise NotImplementedError("Caffe2 is not supported yet.")
'''
assert args.inputShape != None
from dlconv.caffe2.conversion.transformer import Caffe2Transformer
transformer = Caffe2Transformer(args.network, args.weights, args.inputShape, 'tensorflow')
graph = transformer.transform_graph()
data = transformer.transform_data()
from dlconv.common.writer import JsonFormatter, ModelSaver, PyWriter
JsonFormatter(graph).dump(args.dstPath + ".json")
print ("IR saved as [{}.json].".format(args.dstPath))
prototxt = graph.as_graph_def().SerializeToString()
with open(args.dstPath + ".pb", 'wb') as of:
of.write(prototxt)
print ("IR saved as [{}.pb].".format(args.dstPath))
import numpy as np
with open(args.dstPath + ".npy", 'wb') as of:
np.save(of, data)
print ("IR weights saved as [{}.npy].".format(args.dstPath))
return 0
'''
elif args.srcFramework == 'keras':
if args.network != None:
model = (args.network, args.weights)
else:
model = args.weights
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
parser = Keras2Parser(model)
elif args.srcFramework == 'tensorflow':
if args.weights == None:
# only convert network structure
model = args.network
else:
model = (args.network, args.weights)
from mmdnn.conversion.tensorflow.tensorflow_parser import TensorflowParser
parser = TensorflowParser(model, args.dstNodeName)
elif args.srcFramework == 'mxnet':
assert args.inputShape != None
if args.weights == None:
model = (args.network, args.inputShape)
else:
import re
if re.search('.', args.weights):
args.weights = args.weights[:-7]
prefix, epoch = args.weights.rsplit('-', 1)
model = (args.network, prefix, epoch, args.inputShape)
from mmdnn.conversion.mxnet.mxnet_parser import MXNetParser
parser = MXNetParser(model)
else:
raise NotImplementedError("Unknown framework [{}].".format(
args.srcFramework))
parser.gen_IR()
parser.save_to_json(args.dstPath + ".json")
parser.save_to_proto(args.dstPath + ".pb")
parser.save_weights(args.dstPath + ".npy")
return 0
def _convert(args):
if args.inputShape != None:
inputshape = []
for x in args.inputShape:
shape = x.split(',')
inputshape.append([int(x) for x in shape])
else:
inputshape = [None]
if args.srcFramework == 'caffe':
from mmdnn.conversion.caffe.transformer import CaffeTransformer
transformer = CaffeTransformer(args.network,
args.weights,
"tensorflow",
inputshape[0],
phase=args.caffePhase)
graph = transformer.transform_graph()
data = transformer.transform_data()
from mmdnn.conversion.caffe.writer import JsonFormatter, ModelSaver, PyWriter
JsonFormatter(graph).dump(args.dstPath + ".json")
print("IR network structure is saved as [{}.json].".format(
args.dstPath))
prototxt = graph.as_graph_def().SerializeToString()
with open(args.dstPath + ".pb", 'wb') as of:
of.write(prototxt)
print("IR network structure is saved as [{}.pb].".format(args.dstPath))
import numpy as np
with open(args.dstPath + ".npy", 'wb') as of:
np.save(of, data)
print("IR weights are saved as [{}.npy].".format(args.dstPath))
return 0
elif args.srcFramework == 'caffe2':
raise NotImplementedError("Caffe2 is not supported yet.")
elif args.srcFramework == 'keras':
if args.network != None:
model = (args.network, args.weights)
else:
model = args.weights
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
parser = Keras2Parser(model)
elif args.srcFramework == 'tensorflow' or args.srcFramework == 'tf':
if args.dstNodeName is None:
raise ValueError(
"Need to provide the output node of Tensorflow model.")
assert args.network or args.weights
if not args.network:
if args.inNodeName is None:
raise ValueError(
"Need to provide the input node of Tensorflow model.")
if inputshape is None:
raise ValueError(
"Need to provide the input node shape of Tensorflow model."
)
assert len(args.inNodeName) == len(inputshape)
from mmdnn.conversion.tensorflow.tensorflow_frozenparser import TensorflowParser2
parser = TensorflowParser2(args.weights, inputshape,
args.inNodeName, args.dstNodeName)
else:
from mmdnn.conversion.tensorflow.tensorflow_parser import TensorflowParser
if args.inNodeName and inputshape[0]:
parser = TensorflowParser(args.network, args.weights,
args.dstNodeName, inputshape[0],
args.inNodeName)
else:
parser = TensorflowParser(args.network, args.weights,
args.dstNodeName)
elif args.srcFramework == 'mxnet':
assert inputshape != None
if args.weights == None:
model = (args.network, inputshape[0])
else:
import re
if re.search('.', args.weights):
args.weights = args.weights[:-7]
prefix, epoch = args.weights.rsplit('-', 1)
model = (args.network, prefix, epoch, inputshape[0])
from mmdnn.conversion.mxnet.mxnet_parser import MXNetParser
parser = MXNetParser(model)
elif args.srcFramework == 'cntk':
from mmdnn.conversion.cntk.cntk_parser import CntkParser
model = args.network or args.weights
parser = CntkParser(model)
elif args.srcFramework == 'pytorch':
assert inputshape != None
from mmdnn.conversion.pytorch.pytorch_parser import PytorchParser
model = args.network or args.weights
assert model != None
parser = PytorchParser(model, inputshape[0])
elif args.srcFramework == 'torch' or args.srcFramework == 'torch7':
from mmdnn.conversion.torch.torch_parser import TorchParser
model = args.network or args.weights
assert model != None
parser = TorchParser(model, inputshape[0])
elif args.srcFramework == 'onnx':
from mmdnn.conversion.onnx.onnx_parser import ONNXParser
parser = ONNXParser(args.network)
elif args.srcFramework == 'darknet':
from mmdnn.conversion.darknet.darknet_parser import DarknetParser
parser = DarknetParser(args.network, args.weights, args.darknetStart)
elif args.srcFramework == 'coreml':
from mmdnn.conversion.coreml.coreml_parser import CoremlParser
parser = CoremlParser(args.network)
else:
raise ValueError("Unknown framework [{}].".format(args.srcFramework))
parser.run(args.dstPath)
return 0
def _convert(args):
if args.srcFramework == 'caffe':
from mmdnn.conversion.caffe.transformer import CaffeTransformer
transformer = CaffeTransformer(args.network,
args.weights,
"tensorflow",
args.inputShape,
phase=args.caffePhase)
graph = transformer.transform_graph()
data = transformer.transform_data()
from mmdnn.conversion.caffe.writer import JsonFormatter, ModelSaver, PyWriter
JsonFormatter(graph).dump(args.dstPath + ".json")
print("IR network structure is saved as [{}.json].".format(
args.dstPath))
prototxt = graph.as_graph_def().SerializeToString()
with open(args.dstPath + ".pb", 'wb') as of:
of.write(prototxt)
print("IR network structure is saved as [{}.pb].".format(args.dstPath))
import numpy as np
with open(args.dstPath + ".npy", 'wb') as of:
np.save(of, data)
print("IR weights are saved as [{}.npy].".format(args.dstPath))
return 0
elif args.srcFramework == 'caffe2':
raise NotImplementedError("Caffe2 is not supported yet.")
elif args.srcFramework == 'keras':
if args.network != None:
model = (args.network, args.weights)
else:
model = args.weights
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
parser = Keras2Parser(model)
elif args.srcFramework == 'tensorflow' or args.srcFramework == 'tf':
if args.dstNodeName is None:
raise ValueError(
"Need to provide the output node of Tensorflow model.")
# assert args.network or args.frozen_pb
if args.frozen_pb:
from mmdnn.conversion.tensorflow.tensorflow_frozenparser import TensorflowParser2
parser = TensorflowParser2(args.frozen_pb, args.inputShape,
args.dstNodeName)
else:
from mmdnn.conversion.tensorflow.tensorflow_parser import TensorflowParser
parser = TensorflowParser(args.network, args.weights,
args.frozen_pb, args.dstNodeName)
elif args.srcFramework == 'mxnet':
assert args.inputShape != None
if args.weights == None:
model = (args.network, args.inputShape)
else:
import re
if re.search('.', args.weights):
args.weights = args.weights[:-7]
prefix, epoch = args.weights.rsplit('-', 1)
model = (args.network, prefix, epoch, args.inputShape)
from mmdnn.conversion.mxnet.mxnet_parser import MXNetParser
parser = MXNetParser(model)
elif args.srcFramework == 'cntk':
from mmdnn.conversion.cntk.cntk_parser import CntkParser
model = args.network or args.weights
parser = CntkParser(model)
elif args.srcFramework == 'pytorch':
assert args.inputShape != None
from mmdnn.conversion.pytorch.pytorch_parser import PyTorchParser
parser = PyTorchParser(args.network, args.inputShape)
else:
raise ValueError("Unknown framework [{}].".format(args.srcFramework))
parser.run(args.dstPath)
return 0