зеркало из https://github.com/microsoft/MMdnn.git
keras to tensorflow test
This commit is contained in:
Ru Zhang (MSR Student-FA Talent)
Родитель
ea0e40d498
Коммит
60cbea8eaa
|
|
@ -105,7 +105,7 @@ def _main():
|
|||
parser.add_argument(
|
||||
'--path', '-p',
|
||||
type=_text_type,
|
||||
default='test/model/',
|
||||
default='./',
|
||||
help='Path to save the model network file (e.g keras h5')
|
||||
|
||||
args = parser.parse_args()
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ class keras_extractor(base_extractor):
|
|||
|
||||
|
||||
@classmethod
|
||||
def inference(cls, architecture, image_path, model_path='test/model/'):
|
||||
def inference(cls, architecture, image_path):
|
||||
if cls.sanity_check(architecture):
|
||||
model = cls.architecture_map[architecture]()
|
||||
import numpy as np
|
||||
|
|
|
|||
|
|
@ -1,127 +0,0 @@
|
|||
import os
|
||||
import keras
|
||||
import tensorflow as tf
|
||||
import unittest
|
||||
import numpy as np
|
||||
from imp import reload
|
||||
from mmdnn.conversion.examples.imagenet_test import TestKit
|
||||
from mmdnn.conversion.examples.keras.extractor import keras_extractor
|
||||
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
|
||||
from mmdnn.conversion.cntk.cntk_emitter import CntkEmitter
|
||||
from mmdnn.conversion.tensorflow.tensorflow_emitter import TensorflowEmitter
|
||||
|
||||
_Emitter = {
|
||||
'caffe' : "CaffeEmitter",
|
||||
'cntk' : "CntkEmitter",
|
||||
'coreml' : "",
|
||||
'keras' : "Keras2Emitter",
|
||||
'mxnet' : "MXNetEmitter",
|
||||
'pytorch' : "PytorchEmitter",
|
||||
'tensorflow' : "TensorflowEmitter"
|
||||
}
|
||||
image_name = "mmdnn/conversion/examples/data/seagull.jpg"
|
||||
|
||||
def ensure_dir(f):
|
||||
d = os.path.dirname(f)
|
||||
if not os.path.exists(d):
|
||||
os.makedirs(d)
|
||||
|
||||
def _compute_SNR(x,y):
|
||||
noise = x - y
|
||||
noise_var = np.sum(noise ** 2)/len(noise) + 1e-7
|
||||
signal_energy = np.sum(y ** 2)/len(y)
|
||||
max_signal_energy = np.amax(y ** 2)
|
||||
SNR = 10 * np.log10(signal_energy/noise_var)
|
||||
PSNR = 10 * np.log10(max_signal_energy/noise_var)
|
||||
return SNR, PSNR
|
||||
|
||||
def _compute_max_relative_error(x,y):
|
||||
rerror = 0
|
||||
index = 0
|
||||
for i in range(len(x)):
|
||||
den = max(1.0, np.abs(x[i]), np.abs(y[i]))
|
||||
if np.abs(x[i]/den - y[i]/den) > rerror:
|
||||
rerror = np.abs(x[i]/den - y[i]/den)
|
||||
index = i
|
||||
return rerror, index
|
||||
|
||||
class CorrectnessTest(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(self):
|
||||
""" Set up the unit test by loading common utilities.
|
||||
"""
|
||||
self.err_thresh = 0.15
|
||||
self.snr_thresh = 12
|
||||
self.psnr_thresh = 30
|
||||
# self.red_bias = -1
|
||||
# self.blue_bias = -1
|
||||
# self.green_bias = -1
|
||||
# self.image_scale = 2.0/255
|
||||
|
||||
def _compare_outputs(self, keras_out, cntk_out):
|
||||
self.assertEquals(len(keras_out), len(cntk_out))
|
||||
error, ind = _compute_max_relative_error(cntk_out, keras_out)
|
||||
SNR, PSNR = _compute_SNR(cntk_out, keras_out)
|
||||
print("error:", error)
|
||||
print("SNR:", SNR)
|
||||
print("PSNR:", PSNR)
|
||||
self.assertGreater(SNR, self.snr_thresh)
|
||||
self.assertGreater(PSNR, self.psnr_thresh)
|
||||
self.assertLess(error, self.err_thresh)
|
||||
|
||||
class TestModels(CorrectnessTest):
|
||||
def test_keras(self):
|
||||
filename = "test/model/"
|
||||
ensure_dir(filename)
|
||||
# keras original
|
||||
framework_list = ['tensorflow']
|
||||
# network_name_list = ['resnet50','vgg19', 'vgg16','inception_v3']
|
||||
network_name_list = [ 'xception', 'mobilenet']
|
||||
for framework in framework_list:
|
||||
for network_name in network_name_list:
|
||||
original_predict = keras_extractor.inference(network_name, image_name)
|
||||
|
||||
# target framework
|
||||
keras_extractor.download(network_name)
|
||||
model2parser = "test/model/imagenet_{}.h5".format(network_name)
|
||||
|
||||
# to IR
|
||||
parser = Keras2Parser(model2parser)
|
||||
parser.gen_IR()
|
||||
parser.save_to_proto("test/model/" + network_name + "_converted.pb")
|
||||
parser.save_weights("test/model/" + network_name + "_converted.npy")
|
||||
|
||||
# to code
|
||||
Emitter_cls_name = _Emitter[framework]
|
||||
parameter_emitter = '(("test/model/" + network_name + "_converted.pb", "test/model/" + network_name + "_converted.npy"))'
|
||||
emitter = eval(Emitter_cls_name + parameter_emitter)
|
||||
# emitter = CntkEmitter("test/model/" + network_name + "_converted.pb", "test/model/" + network_name + "_converted.npy")
|
||||
emitter.run("converted_model.py", None, 'test')
|
||||
|
||||
|
||||
# import converted model
|
||||
import converted_model
|
||||
reload (converted_model)
|
||||
model_converted = converted_model.KitModel("test/model/" + network_name + "_converted.npy")
|
||||
|
||||
func = TestKit.preprocess_func['keras'][network_name]
|
||||
img = func(image_name)
|
||||
|
||||
if framework == 'cntk':
|
||||
predict = model_converted.eval({model_converted.arguments[0]:[img]})
|
||||
|
||||
elif framework == 'tensorflow':
|
||||
input_tf, model_tf = model_converted
|
||||
input_data = np.expand_dims(img, 0)
|
||||
with tf.Session() as sess:
|
||||
init = tf.global_variables_initializer()
|
||||
sess.run(init)
|
||||
predict = sess.run(model_tf, feed_dict = {input_tf : input_data})
|
||||
|
||||
# print(predict)
|
||||
converted_predict = np.squeeze(predict)
|
||||
self._compare_outputs(original_predict, converted_predict)
|
||||
os.remove("test/model/" + network_name + "_converted.pb")
|
||||
os.remove("test/model/" + network_name + "_converted.npy")
|
||||
os.remove("converted_model.py")
|
||||
|
||||
|
|
@ -2,10 +2,15 @@ import os
|
|||
import unittest
|
||||
import numpy as np
|
||||
from imp import reload
|
||||
import tensorflow as tf
|
||||
from mmdnn.conversion.examples.imagenet_test import TestKit
|
||||
|
||||
from mmdnn.conversion.examples.keras.extractor import keras_extractor
|
||||
|
||||
from mmdnn.conversion.keras.keras2_parser import Keras2Parser
|
||||
|
||||
from mmdnn.conversion.cntk.cntk_emitter import CntkEmitter
|
||||
from mmdnn.conversion.tensorflow.tensorflow_emitter import TensorflowEmitter
|
||||
|
||||
def _compute_SNR(x,y):
|
||||
noise = x - y
|
||||
|
|
@ -28,12 +33,19 @@ def _compute_max_relative_error(x,y):
|
|||
return rerror, index
|
||||
|
||||
|
||||
|
||||
def ensure_dir(f):
|
||||
d = os.path.dirname(f)
|
||||
if not os.path.exists(d):
|
||||
os.makedirs(d)
|
||||
|
||||
|
||||
class CorrectnessTest(unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(self):
|
||||
""" Set up the unit test by loading common utilities.
|
||||
"""
|
||||
self.err_thresh = 0.0015
|
||||
self.err_thresh = 0.15
|
||||
self.snr_thresh = 12
|
||||
self.psnr_thresh = 30
|
||||
|
||||
|
|
@ -96,20 +108,52 @@ class TestModels(CorrectnessTest):
|
|||
return converted_predict
|
||||
|
||||
|
||||
@staticmethod
|
||||
def TensorflowEmit(original_framework, architecture_name, architecture_path, weight_path, image_path):
|
||||
print("Testing {} from {} to Tensorflow.".format(architecture_name, original_framework))
|
||||
|
||||
# IR to code
|
||||
emitter = TensorflowEmitter((architecture_path, weight_path))
|
||||
emitter.run("converted_model.py", None, 'test')
|
||||
del emitter
|
||||
|
||||
# import converted model
|
||||
import converted_model
|
||||
reload (converted_model)
|
||||
model_converted = converted_model.KitModel(TestModels.tmpdir + architecture_name + "_converted.npy")
|
||||
input_tf, model_tf = model_converted
|
||||
|
||||
func = TestKit.preprocess_func[original_framework][architecture_name]
|
||||
img = func(image_path)
|
||||
input_data = np.expand_dims(img, 0)
|
||||
with tf.Session() as sess:
|
||||
init = tf.global_variables_initializer()
|
||||
sess.run(init)
|
||||
predict = sess.run(model_tf, feed_dict = {input_tf : input_data})
|
||||
del model_converted
|
||||
del converted_model
|
||||
os.remove("converted_model.py")
|
||||
converted_predict = np.squeeze(predict)
|
||||
return converted_predict
|
||||
|
||||
|
||||
test_table = {
|
||||
'keras': {
|
||||
'vgg16' : [CntkEmit],
|
||||
'vgg19' : [CntkEmit],
|
||||
'inception_v3' : [CntkEmit],
|
||||
'resnet50' : [CntkEmit],
|
||||
'vgg16' : [CntkEmit,TensorflowEmit],
|
||||
'vgg19' : [CntkEmit,TensorflowEmit],
|
||||
'inception_v3' : [CntkEmit,TensorflowEmit],
|
||||
'resnet50' : [CntkEmit,TensorflowEmit],
|
||||
'densenet' : [CntkEmit],
|
||||
'xception' : [],
|
||||
'xception' : [TensorflowEmit],
|
||||
'mobilenet' : [TensorflowEmit],
|
||||
# 'nasnet' : [],
|
||||
}
|
||||
}
|
||||
|
||||
def test_keras(self):
|
||||
# keras original
|
||||
ensure_dir(self.cachedir)
|
||||
ensure_dir(self.tmpdir)
|
||||
original_framework = 'keras'
|
||||
|
||||
for network_name in self.test_table[original_framework].keys():
|
||||
|
|
|
|||
Загрузка…
Ссылка в новой задаче