def ResNet_cifar( mode='train',root_path='',batch_size=32,block_n=3):
network='name:"ResNet-20"'+'\n'
if mode=='train':
network,last_name=write_prototxt.data(network,name="Data1",mirror=True,scale=0.00390625,crop_size=32,batch_size=batch_size,backend="LMDB",shuffle=True,datasets_path=root_path)
use_global_stats=False
elif mode=='test':
network,last_name=write_prototxt.data(network,name="Data1",mirror=False,scale=0.00390625,crop_size=32,batch_size=batch_size,backend="LMDB",shuffle=False,datasets_path=root_path)
use_global_stats=True
network,last_name=write_prototxt.Convolution(network,name="conv1",bottom_name=last_name,top_name='conv1',num_output=16,
bias_term=True,pad=1,kernel_size=3,stride=1,weight_type='msra',bias_type='constant')
network,last_name=write_prototxt.BatchNorm(network,name_bn="conv1/bn",name_scale="conv1/scale",
bottom_name="conv1",top_name="conv1",use_global_stats=use_global_stats)
network,last_name=write_prototxt.ReLU(network,name="conv1/ReLU",bottom_name='conv1',top_name='conv1')
#network=Pooling(network,name="Pooling1",bottom_name='conv1',top_name='Pooling1',pool='MAX')
network,last_name=ResBlock(network,last_name=last_name,block_name='conv2_',block_n=block_n,num_out=16,downsampling=False,use_global_stats=use_global_stats)
network,last_name=ResBlock(network,last_name=last_name,block_name='conv3_',block_n=block_n,num_out=32,downsampling=True,down_method='conv',use_global_stats=use_global_stats)
network,last_name=ResBlock(network,last_name=last_name,block_name='conv4_',block_n=block_n,num_out=64,downsampling=True,down_method='conv',use_global_stats=use_global_stats)
#network,last_name=ResBlock(network,last_name=last_name,block_name='conv5_',block_n=block_n,num_out=64,downsampling=True,down_method='conv',use_global_stats=use_global_stats)
network,last_name=write_prototxt.Pooling(network,name="Pooling1",bottom_name=last_name,top_name='Pooling1',pool='AVE',global_pooling=True)
network,last_name=write_prototxt.InnerProduct(network,name="fc1",bottom_name=last_name,top_name='fc1',num_output=10,weight_type='msra',bias_type='constant')
#if mode=='train':
network,last_name=write_prototxt.SoftmaxWithLoss(network,name="Softmax1",bottom_name1='fc1',bottom_name2='label',top_name='Softmax1')
if mode=='test':
network,last_name=write_prototxt.Accuracy(network,name="prob",bottom_name1='fc1',bottom_name2='label',top_name='prob')
#
#
print network
return network
def Net( mode='train',root_path='',batch_size=32):
network='name:"MnasNet"'+'\n'
if mode=='train':
network,last_name=write_prototxt.data(network,name="data",mirror=True,scale=0.00390625,crop_size=224,batch_size=batch_size,backend="LMDB",shuffle=True,datasets_path=root_path)
use_global_stats=False
elif mode=='test':
network,last_name=write_prototxt.data(network,name="data",mirror=False,scale=0.00390625,crop_size=224,batch_size=batch_size,backend="LMDB",shuffle=False,datasets_path=root_path)
use_global_stats=True
elif mode=='deploy':
network='input: "data"'+'\n'
network=network+'input_dim: 1'+'\n'
network=network+'input_dim: 3'+'\n'
network=network+'input_dim: 224'+'\n'
network=network+'input_dim: 224'+'\n'
last_name="data"
use_global_stats=True
network,last_name=write_prototxt.Convolution(network,name="conv1",bottom_name=last_name,top_name='conv1',num_output=32,
bias_term=False,pad=1,kernel_size=3,stride=2,weight_type='msra',bias_type='constant')
network,last_name=write_prototxt.BatchNorm(network,name_bn="conv1/bn",name_scale="conv1/scale",
bottom_name="conv1",top_name="conv1",use_global_stats=use_global_stats)
network,last_name=write_prototxt.ReLU(network,name="conv1/ReLU",bottom_name='conv1',top_name='conv1')
network,last_name=MnasNet_block_first(network,last_name=last_name,block_name='conv2_1',
block_n=1,num_out1=32,num_out2=16,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv3_1',
block_n=1,kernel_size=3,num_out1=48,num_out2=24,downsampling=True,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv3_2',
block_n=1,kernel_size=3,num_out1=72,num_out2=24,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv3_3',
block_n=1,kernel_size=3,num_out1=72,num_out2=24,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv4_1',
block_n=1,kernel_size=5,num_out1=72,num_out2=40,downsampling=True,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv4_2',
block_n=1,kernel_size=5,num_out1=120,num_out2=40,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv4_3',
block_n=1,kernel_size=5,num_out1=120,num_out2=40,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv5_1',
block_n=1,kernel_size=5,num_out1=240,num_out2=80,downsampling=True,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv5_2',
block_n=1,kernel_size=5,num_out1=480,num_out2=80,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv5_3',
block_n=1,kernel_size=5,num_out1=480,num_out2=80,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv6_1',
block_n=1,kernel_size=3,num_out1=480,num_out2=96,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv6_2',
block_n=1,kernel_size=3,num_out1=576,num_out2=96,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv7_1',
block_n=1,kernel_size=5,num_out1=576,num_out2=192,downsampling=True,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv7_2',
block_n=1,kernel_size=5,num_out1=1152,num_out2=192,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv7_3',
block_n=1,kernel_size=5,num_out1=1152,num_out2=192,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv7_4',
block_n=1,kernel_size=5,num_out1=1152,num_out2=192,downsampling=False,use_global_stats=use_global_stats)
network,last_name=MnasNet_block_bdce(network,last_name=last_name,block_name='conv8_1',
block_n=1,kernel_size=3,num_out1=1152,num_out2=320,downsampling=False,use_global_stats=use_global_stats)
network,last_name=write_prototxt.Pooling(network,name="Pooling1",bottom_name=last_name,top_name='Pooling1',pool='AVE',global_pooling=True)
network,last_name=write_prototxt.InnerProduct(network,name="fc1",bottom_name=last_name,top_name='fc1',num_output=2,weight_type='msra',bias_type='constant')
#if mode=='train':
network,last_name=write_prototxt.SoftmaxWithLoss(network,name="Softmax1",bottom_name1='fc1',bottom_name2='label',top_name='Softmax1')
if mode=='test':
network,last_name=write_prototxt.Accuracy(network,name="prob",bottom_name1='fc1',bottom_name2='label',top_name='prob')
#if mode=='deploy':
#network,last_name=write_prototxt.Softmax(network,name="prob",bottom_name='fc1',top_name='prob')
# print network
return network
def Net(mode='train', root_path='', batch_size=32):
network = 'name:"MobileNetV1"' + '\n'
if mode == 'train':
network, last_name = write_prototxt.data(network,
name="Data1",
mirror=True,
crop_size=28,
batch_size=batch_size,
backend="LMDB",
shuffle=True,
datasets_path=root_path)
use_global_stats = False
elif mode == 'test':
network, last_name = write_prototxt.data(network,
name="Data1",
mirror=False,
crop_size=28,
batch_size=batch_size,
backend="LMDB",
shuffle=False,
datasets_path=root_path)
use_global_stats = True
network, last_name = write_prototxt.Convolution(network,
name="conv1",
bottom_name=last_name,
top_name='conv1',
num_output=32,
bias_term=False,
pad=1,
kernel_size=3,
stride=1,
weight_type='msra',
bias_type='constant')
network, last_name = write_prototxt.BatchNorm(
network,
name_bn="conv1_bn",
name_scale="conv1__scale",
bottom_name="conv1",
top_name="conv1",
use_global_stats=use_global_stats)
network, last_name = write_prototxt.ReLU(network,
name="conv1_ReLU",
bottom_name='conv1',
top_name='conv1')
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv2_',
block_n=1,
num_out1=32,
num_out2=64,
downsampling=False,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv3_',
block_n=1,
num_out1=64,
num_out2=128,
downsampling=True,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv4_',
block_n=1,
num_out1=128,
num_out2=128,
downsampling=False,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv5_',
block_n=1,
num_out1=128,
num_out2=256,
downsampling=False,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv6_',
block_n=1,
num_out1=256,
num_out2=512,
downsampling=True,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv7_',
block_n=1,
num_out1=512,
num_out2=512,
downsampling=False,
use_global_stats=use_global_stats)
network, last_name = Mobile_Unit(network,
last_name=last_name,
block_name='conv8_',
block_n=1,
num_out1=512,
num_out2=512,
downsampling=False,
use_global_stats=use_global_stats)
network, last_name = write_prototxt.Pooling(network,
name="Pooling1",
bottom_name=last_name,
top_name='Pooling1',
pool='AVE',
global_pooling=True)
network, last_name = write_prototxt.InnerProduct(network,
name="fc1",
bottom_name=last_name,
top_name='fc1',
num_output=10,
weight_type='xavier',
bias_type='constant')
if mode == 'train':
network, last_name = write_prototxt.SoftmaxWithLoss(
network,
name="Softmax1",
bottom_name1='fc1',
bottom_name2='Data2',
top_name='Softmax1')
if mode == 'test':
network, last_name = write_prototxt.Accuracy(network,
name="prob",
bottom_name1='fc1',
bottom_name2='Data2',
top_name='prob')
#
#
print network
return network
def Net(mode='train', root_path='', batch_size=32):
network = 'name:"ShuffleNet"' + '\n'
if mode == 'train':
network, last_name = write_prototxt.data(network,
name="Data1",
mirror=True,
crop_size=28,
batch_size=batch_size,
backend="LMDB",
shuffle=True,
datasets_path=root_path)
use_global_stats = False
elif mode == 'test':
network, last_name = write_prototxt.data(network,
name="Data1",
mirror=False,
crop_size=28,
batch_size=batch_size,
backend="LMDB",
shuffle=False,
datasets_path=root_path)
use_global_stats = True
network, last_name = write_prototxt.Convolution(network,
name="conv1",
bottom_name=last_name,
top_name='conv1',
num_output=15,
bias_term=False,
pad=1,
kernel_size=3,
stride=1,
weight_type='msra',
bias_type='constant')
network, last_name = write_prototxt.BatchNorm(
network,
name_bn="conv1_bn",
name_scale="conv1__scale",
bottom_name="conv1",
top_name="conv1",
use_global_stats=use_global_stats)
network, last_name = write_prototxt.ReLU(network,
name="conv1_ReLU",
bottom_name='conv1',
top_name='conv1')
network, last_name = ShuffleNet_Unit0(network,
last_name=last_name,
block_name='resx1_conv',
block_n=1,
num_out0=24,
num_out1=45,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx2_conv',
block_n=1,
num_out0=30,
num_out1=60,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx3_conv',
block_n=1,
num_out0=30,
num_out1=60,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx4_conv',
block_n=1,
num_out0=30,
num_out1=60,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx5_conv',
block_n=1,
num_out0=30,
num_out1=60,
downsampling=True,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx6_conv',
block_n=1,
num_out0=48,
num_out1=120,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx7_conv',
block_n=1,
num_out0=48,
num_out1=120,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx10_conv',
block_n=1,
num_out0=48,
num_out1=120,
downsampling=True,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx11_conv',
block_n=1,
num_out0=60,
num_out1=240,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx12_conv',
block_n=1,
num_out0=60,
num_out1=240,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = ShuffleNet_Unit(network,
last_name=last_name,
block_name='resx13_conv',
block_n=1,
num_out0=60,
num_out1=240,
downsampling=False,
group=3,
use_global_stats=use_global_stats)
network, last_name = write_prototxt.Pooling(network,
name="Pooling1",
bottom_name=last_name,
top_name='Pooling1',
pool='AVE',
global_pooling=True)
network, last_name = write_prototxt.InnerProduct(network,
name="fc1",
bottom_name=last_name,
top_name='fc1',
num_output=10,
weight_type='xavier',
bias_type='constant')
if mode == 'train':
network, last_name = write_prototxt.SoftmaxWithLoss(
network,
name="Softmax1",
bottom_name1='fc1',
bottom_name2='Data2',
top_name='Softmax1')
if mode == 'test':
network, last_name = write_prototxt.Accuracy(network,
name="prob",
bottom_name1='fc1',
bottom_name2='Data2',
top_name='prob')
#
#
print network
return network