def convert_kerastf_2_script(model_path=None, out_format=None):
"""
Convert model from keras.model to a script that allows to create the model
using pytorch
model_path - string, path to keras.model. model_path should point to a copy
of the model in the temp folder to make sure the relative path does not
contain spaces " ". Spaces cause isses due to a command-line interface of mmdnn.
outpath
out_format: string, indicating the target format. This can be "tensorflow","pytorch", "caffe","cntk","mxnet","onnx"
"""
tf.reset_default_graph() #Make sure to start with a fresh session
sess = tf.InteractiveSession()
temp_path = aid_bin.create_temp_folder(
) #create a temp folder if it does not already exist
#Create a random filename for a temp. file
tmp_model = np.random.choice(list("STERNBURGPILS"), 5, replace=True)
tmp_model = "".join(tmp_model) + ".model"
tmp_model = os.path.join(temp_path, tmp_model)
shutil.copyfile(model_path, tmp_model) #copy the original model file there
#Get the relative path to temp to make sure there are no spaces " " in the path
relpath = os.path.relpath(tmp_model, start=os.curdir)
#Keras to intermediate representation (IR): convert the temp. file
parser = convertToIR._get_parser()
dstPath = os.path.relpath(tmp_model, start=os.curdir)
dstPath = os.path.splitext(dstPath)[0] #remove the .model file extension
string = "--srcFramework keras --weights " + relpath + " --dstPath " + dstPath
args = parser.parse_args(string.split())
convertToIR._convert(args)
#IR to Code
dstModelPath = dstPath + ".py"
dstWeightPath = dstPath + ".npy"
IRModelPath = dstPath + ".pb"
parser = IRToCode._get_parser()
string = "--dstFramework " + out_format + " --dstModelPath " + dstModelPath + " --IRModelPath " + IRModelPath + " --IRWeightPath " + dstWeightPath + " --dstWeightPath " + dstWeightPath
args = parser.parse_args(string.split())
IRToCode._convert(args)
#Copy the final output script and weights back to the original folder
out_script = os.path.splitext(
model_path)[0] + ".py" #remove .model, put .py instead
out_weights = os.path.splitext(
model_path)[0] + ".npy" #remove .model, put .py instead
shutil.copyfile(dstModelPath, out_script) #copy from temp to original path
shutil.copyfile(dstWeightPath,
out_weights) #copy from temp to original path
#delete all the temp. files
del_json = os.path.splitext(relpath)[0] + ".json"
for file in [IRModelPath, dstModelPath, dstWeightPath, relpath, del_json]:
try:
os.remove(file)
print("temp. file deleted: " + file)
except:
print("temp. file not found/ not deleted: " + file)
sess.close()
def _main():
parser = _get_parser()
args, unknown_args = parser.parse_known_args()
temp_filename = uuid.uuid4().hex
ir_args, unknown_args = _extract_ir_args(args, unknown_args, temp_filename)
ret = convertToIR._convert(ir_args)
if int(ret) != 0:
_sys.exit(int(ret))
if args.dstFramework != 'coreml':
network_filename = get_network_filename(args.dstFramework,
temp_filename,
args.outputModel)
code_args, unknown_args = _extract_code_args(args, unknown_args,
temp_filename,
network_filename)
ret = IRToCode._convert(code_args)
if int(ret) != 0:
_sys.exit(int(ret))
from mmdnn.conversion._script.dump_code import dump_code
dump_code(args.dstFramework, network_filename + '.py',
temp_filename + '.npy', args.outputModel, args.dump_tag)
#remove_temp_files(temp_filename)
else:
model_args, unknown_args = _extract_model_args(args, unknown_args,
temp_filename)
ret = IRToModel._convert(model_args)
#remove_temp_files(temp_filename)
_sys.exit(int(ret))
def _extract_code_args(args, unknown_args, temp_filename, network_filename):
unknown_args.extend(['--dstFramework', args.dstFramework])
unknown_args.extend(['--IRModelPath', temp_filename + '.pb'])
unknown_args.extend(['--IRWeightPath', temp_filename + '.npy'])
unknown_args.extend(['--dstModelPath', network_filename + '.py'])
unknown_args.extend(['--dstWeightPath', temp_filename + '.npy'])
code_parser = IRToCode._get_parser()
return code_parser.parse_known_args(unknown_args)
def _extract_code_args(args, unknown_args, temp_filename):
unknown_args.extend(['--dstFramework', args.dstFramework])
unknown_args.extend(['--IRModelPath', temp_filename + '.pb'])
unknown_args.extend(['--IRWeightPath', temp_filename + '.npy'])
unknown_args.extend(['--dstModelPath', temp_filename + '.py'])
unknown_args.extend(['--dstWeightPath', temp_filename + '.npy'])
code_parser = IRToCode._get_parser()
return code_parser.parse_known_args(unknown_args)
def _main():
parser = _get_parser()
args, unknown_args = parser.parse_known_args()
temp_filename = uuid.uuid4().hex
ir_args, unknown_args = _extract_ir_args(args, unknown_args, temp_filename)
ret = convertToIR._convert(ir_args)
if int(ret) != 0:
_sys.exit(int(ret))
if args.dstFramework != 'coreml':
code_args, unknown_args = _extract_code_args(args, unknown_args, temp_filename)
ret = IRToCode._convert(code_args)
if int(ret) != 0:
_sys.exit(int(ret))
from mmdnn.conversion._script.dump_code import dump_code
dump_code(args.dstFramework, temp_filename + '.py', temp_filename + '.npy', args.outputModel)
remove_temp_files(temp_filename)
elif args.dstType == 'model':
model_args, unknown_args = _extract_model_args(args, unknown_args, temp_filename)
ret = IRToModel._convert(model_args)
remove_temp_files(temp_filename)
_sys.exit(int(ret))