def strides():
"""
Tests an example CNN with additional unusual strides.
:return: None.
"""
num_imgs = 10
nncg = NNCG()
strides = Sequential()
strides.add(
Convolution2D(4, (3, 3),
input_shape=(101, 101, 1),
activation='relu',
padding='same',
strides=(3, 3)))
strides.add(MaxPooling2D(pool_size=(2, 2)))
strides.add(
Convolution2D(8, (3, 3),
padding='valid',
activation='relu',
strides=(2, 3)))
strides.add(Convolution2D(16, (3, 3), padding='valid', activation='relu'))
strides.add(Flatten())
strides.add(Dense(2, activation='softmax'))
images = random_imdb(num_imgs, strides.input.shape[1:].as_list())
nncg.keras_compile(images, strides, 'strides.c')
print_success('strides')
def dense_model():
"""
Tests an example CNN with a Dense layer and valid padding.
:return: None.
"""
num_imgs = 10
nncg = NNCG()
dense_model = Sequential()
dense_model.add(
Convolution2D(8, (3, 3),
input_shape=(70, 50, 1),
activation='relu',
padding='same'))
dense_model.add(MaxPooling2D(pool_size=(2, 2)))
dense_model.add(
Convolution2D(16, (3, 3),
padding='valid',
activation='relu',
bias_initializer='random_uniform'))
dense_model.add(MaxPooling2D(pool_size=(2, 2)))
dense_model.add(
Convolution2D(32, (3, 3),
padding='valid',
activation='relu',
bias_initializer='random_uniform'))
dense_model.add(MaxPooling2D(pool_size=(2, 2)))
dense_model.add(Dropout(0.4))
dense_model.add(Flatten())
dense_model.add(Dense(2, activation='softmax'))
images = random_imdb(num_imgs, dense_model.input.shape[1:].as_list())
nncg.keras_compile(images, dense_model, 'dense_model.c')
print_success('dense_model')
def no_dense():
"""
Tests an example CNN with no dense layer.
:return: None
"""
num_imgs = 10
nncg = NNCG()
no_dense = Sequential()
no_dense.add(
Convolution2D(4, (3, 3),
input_shape=(36, 18, 1),
activation='relu',
padding='same'))
no_dense.add(MaxPooling2D(pool_size=(2, 2)))
no_dense.add(
Convolution2D(8, (3, 3),
padding='same',
activation='relu',
bias_initializer='random_uniform'))
no_dense.add(MaxPooling2D(pool_size=(2, 2)))
no_dense.add(
Convolution2D(16, (3, 3),
padding='same',
activation='relu',
bias_initializer='random_uniform')) # Could be softmax
no_dense.add(MaxPooling2D(pool_size=(4, 2)))
no_dense.add(Dropout(0.4))
no_dense.add(Convolution2D(2, (2, 2), activation='softmax'))
no_dense.add(Flatten())
images = random_imdb(num_imgs, no_dense.input.shape[1:].as_list())
nncg.keras_compile(images, no_dense, 'no_dense.c')
print_success('no_dense')
def VGG19_test():
"""
Tests a full VGG19.
:return: None.
"""
num_imgs = 1
nncg = NNCG()
vgg19_m = VGG19(weights=None)
db = random_imdb(num_imgs, vgg19_m.input.shape[1:].as_list())
nncg.keras_compile(db, vgg19_m, 'vgg19.c', weights_method='stdio')
print_success('VGG19')
def VGG19_quantized_test(db):
"""
Tests a full VGG19 using quantization.
:param db: Image database to test with real images.
:return: None.
"""
nncg = NNCG()
vgg19_m = VGG19(weights='imagenet')
nncg.keras_compile(db,
vgg19_m,
'vgg19.c',
weights_method='stdio',
quatization=True,
arch='sse3',
test_mode='classification')
print_success('VGG19')
model_path = "model.h5"
code_path = "."
if args.imgdb_path is not None:
imgdb_path = args.imgdb_path
if args.model_path is not None:
model_path = args.model_path
if args.code_path is not None:
code_path = args.code_path
images = load_imdb(imgdb_path)
model = load_model(model_path, compile=False)
general_generator = NNCG()
general_generator.keras_compile(images["images"],
model,
code_path,
"qsse3",
arch="sse3",
testing=1000,
quatization=True,
test_mode='classification')
sse_generator = NNCG()
sse_generator.keras_compile(images["images"],
model,
code_path,
"sse3",
arch="sse3",
testing=1000)
parser.add_argument('-b', '--database-path', dest='imgdb_path',
help='Path to the image database to use for training. '
'Default is img.db in current folder.')
parser.add_argument('-m', '--model-path', dest='model_path',
help='Store the trained model using this path. Default is model.h5.')
parser.add_argument('-c', '--code-path', dest='code_path',
help='Path where the file is to be stored. Default is current directory')
args = parser.parse_args()
imgdb_path = "img.db"
model_path = "model.h5"
code_path = "."
if args.imgdb_path is not None:
imgdb_path = args.imgdb_path
if args.model_path is not None:
model_path = args.model_path
if args.code_path is not None:
code_path = args.code_path
images = load_imdb(imgdb_path)
model = load_model(model_path, compile=False)
general_generator = NNCG()
general_generator.keras_compile(images["images"], model, code_path, "gen", arch="general", testing=1000)
sse_generator = NNCG()
sse_generator.keras_compile(images["images"], model, code_path, "sse3", arch="sse3", testing=1000)