def get_model(model_type):
if model_type == 'shufflenet':
model = ShuffleNet(groups=3, weights=None)
elif model_type == 'shufflenet_v2':
model = ShuffleNetV2(bottleneck_ratio=1, weights=None)
else:
raise ValueError('Unsupported model type')
return model
def get_model(model_type, include_top=True):
if model_type == 'shufflenet':
input_shape = (224, 224, 3)
model = ShuffleNet(groups=3, weights=None, include_top=include_top)
elif model_type == 'shufflenet_v2':
input_shape = (224, 224, 3)
model = ShuffleNetV2(bottleneck_ratio=1, weights=None, include_top=include_top)
elif model_type == 'nanonet':
input_shape = (224, 224, 3)
model = NanoNet(weights=None, include_top=include_top)
else:
raise ValueError('Unsupported model type')
return model, input_shape[:2]
def get_model(model_type, include_top=True):
if model_type == 'shufflenet':
input_shape = (224, 224, 3)
model = ShuffleNet(input_shape=input_shape,
groups=3,
weights=None,
include_top=include_top)
elif model_type == 'shufflenet_v2':
input_shape = (224, 224, 3)
model = ShuffleNetV2(input_shape=input_shape,
bottleneck_ratio=1,
weights=None,
include_top=include_top)
elif model_type == 'nanonet':
input_shape = (224, 224, 3)
model = NanoNet(input_shape=input_shape,
weights=None,
include_top=include_top)
elif model_type == 'darknet53':
input_shape = (224, 224, 3)
model = DarkNet53(input_shape=input_shape,
weights=None,
include_top=include_top)
elif model_type == 'cspdarknet53':
input_shape = (224, 224, 3)
model = CSPDarkNet53(input_shape=input_shape,
weights=None,
include_top=include_top)
elif model_type == 'mobilevit_s':
input_shape = (256, 256, 3)
model = MobileViT_S(input_shape=input_shape,
weights=None,
include_top=include_top)
elif model_type == 'mobilevit_xs':
input_shape = (256, 256, 3)
model = MobileViT_XS(input_shape=input_shape,
weights=None,
include_top=include_top)
elif model_type == 'mobilevit_xxs':
input_shape = (256, 256, 3)
model = MobileViT_XXS(input_shape=input_shape,
weights=None,
include_top=include_top)
else:
raise ValueError('Unsupported model type')
return model, input_shape[:2]
def MyNet(input_shape=(218, 178, 3), n_classes=2360):
img_input = Input(shape=input_shape)
# Stem block: 52 x 42 x 64
x = conv2d_bn(img_input, 16, 3, strides=2, padding='valid')
x = conv2d_bn(x, 16, 3, padding='valid')
x = conv2d_bn(x, 32, 3)
x = MaxPooling2D(3, strides=2)(x)
# Mixed 5b (Inception-A block): 52 x 42 x 160
branch_0 = conv2d_bn(x, 48, 1)
branch_1 = conv2d_bn(x, 24, 1)
branch_1 = conv2d_bn(branch_1, 32, 5)
branch_2 = conv2d_bn(x, 32, 1)
branch_2 = conv2d_bn(branch_2, 48, 3)
branch_2 = conv2d_bn(branch_2, 48, 3)
branch_pool = AveragePooling2D(3, strides=1, padding='same')(x)
branch_pool = conv2d_bn(branch_pool, 32, 1)
branches = [branch_0, branch_1, branch_2, branch_pool]
channel_axis = 1 if K.image_data_format() == 'channels_first' else 3
x = Concatenate(axis=channel_axis, name='mixed_5b')(branches)
# 5x block35 (Inception-ResNet-A block): 52 x 42 x 160
for block_idx in range(1, 6):
x = separable_inception_resnet_block(x,
scale=0.17,
block_type='block35',
block_idx=block_idx)
shuffle_model = ShuffleNet(include_top=True,
input_tensor=x,
scale_factor=1.0,
num_shuffle_units=[10, 5],
groups=1,
pooling='avg',
classes=n_classes)
# Create model
model = Model(img_input, shuffle_model.output, name='MyNet')
model.summary()
return model
def preprocess(x):
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
x /= 255.0
x -= 0.5
x *= 2.0
return x
if __name__ == '__main__':
groups = 3
batch_size = 128
inital_epoch = 0
ds = '/mnt/daten/Development/ILSVRC2012_256'
model = ShuffleNet(groups=groups, pooling='avg')
plot_model(model, 'model.png', show_shapes=True)
# model.load_weights('%s.hdf5' % model.name, by_name=True)
csv_logger = CSVLogger('%s.log' % model.name, append=(inital_epoch is not 0))
checkpoint = ModelCheckpoint(filepath='%s.hdf5' % model.name, verbose=1,
save_best_only=True, monitor='val_acc', mode='max')
learn_rates = [0.05, 0.01, 0.005, 0.001, 0.0005]
lr_scheduler = LearningRateScheduler(lambda epoch: learn_rates[epoch // 30])
train_datagen = ImageDataGenerator(preprocessing_function=preprocess,
zoom_range=0.25,
width_shift_range=0.05,
height_shift_range=0.05,
horizontal_flip=True)
from shufflenet import ShuffleNet model = ShuffleNet()