def __init__(self, layers=16, use_bn=False, num_classes=1000):
super(VGG, self).__init__()
self.layers = layers
self.use_bn = use_bn
supported_layers = [16, 19]
assert layers in supported_layers
if layers == 16:
depth = [2, 2, 3, 3, 3]
elif layers == 19:
depth = [2, 2, 4, 4, 4]
num_channels = [3, 64, 128, 256, 512]
num_filters = [64, 128, 256, 512, 512]
self.layer1 = fluid.dygraph.Sequential(*self.make_layer(num_channels[0], num_filters[0], depth[0], use_bn, name='layer1'))
self.layer2 = fluid.dygraph.Sequential(*self.make_layer(num_channels[1], num_filters[1], depth[1], use_bn, name='layer2'))
self.layer3 = fluid.dygraph.Sequential(*self.make_layer(num_channels[2], num_filters[2], depth[2], use_bn, name='layer3'))
self.layer4 = fluid.dygraph.Sequential(*self.make_layer(num_channels[3], num_filters[3], depth[3], use_bn, name='layer4'))
self.layer5 = fluid.dygraph.Sequential(*self.make_layer(num_channels[4], num_filters[4], depth[4], use_bn, name='layer5'))
self.classifier = fluid.dygraph.Sequential(
Linear(input_dim=512 * 7 * 7, output_dim=4096, act='relu'),
Dropout(),
Linear(input_dim=4096, output_dim=4096, act='relu'),
Dropout(),
Linear(input_dim=4096, output_dim=num_classes))
self.out_dim = 512 * 7 * 7
def __init__(self, num_classes=59, backbone='resnet50'):
super(PSPNet, self).__init__()
res = ResNet50(pretrained=False)
# stem: res.conv, res.pool2d_max
self.layer0 = fluid.dygraph.Sequential(
res.conv,
res.pool2d_max
)
self.layer1 = res.layer1
self.layer2 = res.layer2
self.layer3 = res.layer3
self.layer4 = res.layer4
num_channels = 2048
# psp: 2048 -> 2048*2
self.pspmodule = PSPModule(num_channels, [1, 2, 3, 6])
num_channels *= 2
# cls: 2048*2 -> 512 -> num_classes
self.classifier = fluid.dygraph.Sequential(
Conv2D(num_channels, num_filters=512, filter_size=3, padding=1),
BatchNorm(512, act='relu'),
Dropout(0.1),
Conv2D(512, num_classes, filter_size=1)
)
def __init__(self, num_classes=1):
super(CNN_AllTricks, self).__init__()
self.conv1 = Conv2D(3, 64, 3, padding=1, stride=1, act='leaky_relu')
self.bn1 = BatchNorm(64)
self.conv2 = Conv2D(64, 128, 3, padding=1, stride=1, act='leaky_relu')
self.bn2 = BatchNorm(128)
self.conv3 = Conv2D(128, 256, 3, padding=1, stride=1, act='leaky_relu')
self.bn3 = BatchNorm(256)
self.conv4 = Conv2D(256, 512, 3, padding=1, stride=1, act='leaky_relu')
self.bn4 = BatchNorm(512)
self.block5 = ConcatConv((512, 384), (384, 256), (256, 256),
128,
act_fun='leaky_relu')
self.bn5 = BatchNorm(1024)
self.block6 = ConcatConv((1024, 384), (384, 256), (256, 256),
128,
act_fun='leaky_relu')
self.bn6 = BatchNorm(1024)
self.pool_global = Pool2D(pool_stride=1,
global_pooling=True,
pool_type='avg')
self.fc = Linear(input_dim=1024, output_dim=num_classes)
self.pool_down = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.dropout = Dropout(p=0.5)
def __init__(self,
block=BasicBlock,
layers=50,
inp=3,
num_classes=400,
input_size=112,
dropout=0.5):
self.inplanes = 64
self.inp = inp
super(ResNet, self).__init__()
self.conv1 = Conv2D(inp,
64,
filter_size=7,
stride=2,
padding=3,
bias_attr=False)
self.bn1 = BatchNorm(64)
self.relu = fluid.layers.relu #nn.ReLU(inplace=True)
self.maxpool = Pool2D(
pool_size=3, pool_stride=2, pool_padding=1,
pool_type='max') #nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.rep_of_rep = repofrep("flowofflow")
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
# probably need to adjust this based on input spatial size
size = int(math.ceil(input_size / 32))
self.avgpool = Pool2D(pool_size=size,
pool_stride=1,
pool_padding=0,
pool_type='avg') #nn.AvgPool2d(size, stride=1)
self.dropout = Dropout(dropout) #nn.Dropout(p=dropout)
self.fc = Linear(512 * block.expansion, num_classes)
def __init__(self, num_classes=1):
super(AlexNet, self).__init__()
self.conv1 = Conv2D(num_channels=3, num_filters=96, filter_size=11, stride=4, padding=2, act='relu')
self.pool1 = Pool2D(pool_size=3, pool_stride=2, pool_type='max')
self.conv2 = Conv2D(num_channels=96, num_filters=256, filter_size=5, stride=1, padding=2, act='relu')
self.pool2 = Pool2D(pool_size=3, pool_stride=2, pool_type='max')
self.conv3 = Conv2D(num_channels=256, num_filters=384, filter_size=3, stride=1, padding=1, act='relu')
self.conv4 = Conv2D(num_channels=384, num_filters=384, filter_size=3, stride=1, padding=1, act='relu')
self.conv5 = Conv2D(num_channels=384, num_filters=256, filter_size=3, stride=1, padding=1, act='relu')
self.pool5 = Pool2D(pool_size=3, pool_stride=2, pool_type='max')
self.fc1 = Linear(256 * 6 * 6, 4096, act='relu')
self.drop_out1 = Dropout(p=0.5)
self.fc2 = Linear(4096, 4096, act='relu')
self.drop_out2 = Dropout(p=0.5)
self.fc3 = Linear(4096, num_classes, act='relu')
def _prepare_tsn(self, num_class):
feature_dim = getattr(self.base_model,
self.base_model.last_layer_name).weight.shape[0]
setattr(self.base_model, self.base_model.last_layer_name,
Dropout(p=self.dropout))
self.new_fc = Linear(feature_dim,
num_class,
act='softmax',
param_attr=fluid.ParamAttr(learning_rate=1.0),
bias_attr=fluid.ParamAttr(learning_rate=2.0))
def __init__(self, hidden_unit_num, image_size, in_channel, patch_num,
dropout):
super(ResNet24, self).__init__()
self.patch_size = image_size // patch_num
self.patch_num = patch_num
n_patch_size = (image_size // patch_num) * (image_size // patch_num)
self.position_embedding = fluid.layers.create_parameter(
(1, n_patch_size + 1, hidden_unit_num),
dtype='float32',
is_bias=True)
self.cls_token = fluid.layers.create_parameter((1, 1, hidden_unit_num),
is_bias=False,
dtype='float32')
self.dropout = Dropout(dropout)
self.layer1_1 = BasicBlock(num_channels=in_channel,
num_filters=24,
stride=1,
shortcut=False)
self.layer1_2 = BasicBlock(num_channels=24,
num_filters=48,
stride=2,
shortcut=False)
self.layer2_1 = BasicBlock(num_channels=48,
num_filters=48,
stride=1,
shortcut=False)
self.layer2_2 = BasicBlock(num_channels=48,
num_filters=48,
stride=2,
shortcut=False)
self.layer3_1 = BasicBlock(num_channels=48,
num_filters=128,
stride=1,
shortcut=False)
self.layer3_2 = BasicBlock(num_channels=128,
num_filters=256,
stride=2,
shortcut=False)
self.layer4_1 = BasicBlock(num_channels=256,
num_filters=256,
stride=1,
shortcut=False)
self.layer4_2 = BasicBlock(num_channels=256,
num_filters=256,
stride=1,
shortcut=False)
self.layer5_1 = BasicBlock(num_channels=256,
num_filters=512,
stride=1,
shortcut=False)
self.layer5_2 = BasicBlock(num_channels=512,
num_filters=hidden_unit_num,
stride=2,
shortcut=False)
def __init__(self, num_classes=59):
super(FCN8s, self).__init__()
backbone = VGG16BN(pretrained=False)
self.layer1 = backbone.layer1
self.layer1[0].conv._padding = [100, 100]
self.pool1 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer2 = backbone.layer2
self.pool2 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer3 = backbone.layer3
self.pool3 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer4 = backbone.layer4
self.pool4 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer5 = backbone.layer5
self.pool5 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.fc6 = Conv2D(512, 4096, 7, act='relu')
self.fc7 = Conv2D(4096, 4096, 1, act='relu')
self.drop6 = Dropout()
self.drop7 = Dropout()
self.score = Conv2D(4096, num_classes, 1)
self.score_pool3 = Conv2D(256, num_classes, 1)
self.score_pool4 = Conv2D(512, num_classes, 1)
self.up_output = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=4,
stride=2,
bias_attr=False)
self.up_pool4 = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=4,
stride=2,
bias_attr=False)
self.up_final = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=16,
stride=8,
bias_attr=False)
def __init__(self, name_scope, out_chs=20, in_chs=1024, inter_chs=512):
super(DANet, self).__init__(name_scope)
name_scope = self.full_name()
self.in_chs = in_chs
self.out_chs = out_chs
self.inter_chs = inter_chs if inter_chs else in_chs
self.backbone = ResNet(50)
self.conv5p = Sequential(
Conv2D(self.in_chs, self.inter_chs, 3, padding=1),
BatchNorm(self.inter_chs, act='relu'),
)
self.conv5c = Sequential(
Conv2D(self.in_chs, self.inter_chs, 3, padding=1),
BatchNorm(self.inter_chs, act='relu'),
)
self.sp = PAM_module(self.inter_chs)
self.sc = CAM_module(self.inter_chs)
self.conv6p = Sequential(
Conv2D(self.inter_chs, self.inter_chs, 3, padding=1),
BatchNorm(self.inter_chs, act='relu'),
)
self.conv6c = Sequential(
Conv2D(self.inter_chs, self.inter_chs, 3, padding=1),
BatchNorm(self.inter_chs, act='relu'),
)
self.conv7p = Sequential(
Dropout(0.1),
Conv2D(self.inter_chs, self.out_chs, 1),
)
self.conv7c = Sequential(
Dropout(0.1),
Conv2D(self.inter_chs, self.out_chs, 1),
)
self.conv7pc = Sequential(
Dropout(0.1),
Conv2D(self.inter_chs, self.out_chs, 1),
)
def __init__(self,num_classes=59):
super(FCN8s,self).__init__()
vgg16bn = VGG16BN()
self.layer1 = vgg16bn.layer1
self.layer1[0].conv._padding = [100,100]
self.layer2 = vgg16bn.layer2
self.layer3 = vgg16bn.layer3
self.layer4 = vgg16bn.layer4
self.layer5 = vgg16bn.layer5
# self.conv1_1 = Conv2D(3,64,3,padding=1)
# self.conv1_2 = Conv2D(64,64,3,padding=1)
self.pool1 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv2_1 = Conv2D(64,128,3,padding=1)
# self.conv2_2 = Conv2D(128,128,3,padding=1)
self.pool2 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv3_1 = Conv2D(128,256,3,padding=1)
# self.conv3_2 = Conv2D(256,256,3,padding=1)
# self.conv3_3 = Conv2D(256,256,3,padding=1)
self.pool3 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv4_1 = Conv2D(256,512,3,padding=1)
# self.conv4_2 = Conv2D(512,512,3,padding=1)
# self.conv4_3 = Conv2D(512,512,3,padding=1)
self.pool4 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv5_1 = Conv2D(512,512,3,padding=1)
# self.conv5_2 = Conv2D(512,512,3,padding=1)
# self.conv5_3 = Conv2D(512,512,3,padding=1)
self.pool5 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
self.conv6 = Conv2D(512,4096,1,act='relu')
self.conv7 = Conv2D(4096,4096,1,act='relu')
self.drop6 = Dropout()
self.drop7 = Dropout()
self.score = Conv2D(4096,num_classes,1)
self.score_pool3 = Conv2D(256,num_classes,1)
self.score_pool4 = Conv2D(512,num_classes,1,)
self.upsample1 = Conv2DTranspose(num_classes,num_classes,filter_size=4,stride=2,padding=2,bias_attr=False)
self.upsample2 = Conv2DTranspose(num_classes,num_classes,filter_size=4,stride=2,padding=2,bias_attr=False)
self.upsample3 = Conv2DTranspose(num_classes,num_classes,filter_size=16,stride=8,padding=1,bias_attr=False)
def __init__(self, num_classes=1):
super(GoogLeNet_BN, self).__init__()
self.bn64 = BatchNorm(64)
self.bn192 = BatchNorm(192)
self.bn256 = BatchNorm(256)
self.bn480 = BatchNorm(480)
self.bn512 = BatchNorm(512)
self.bn528 = BatchNorm(528)
self.bn832 = BatchNorm(832)
self.bn1024 = BatchNorm(1024)
self.conv1 = Conv2D(3, 64, 7, padding=3, stride=2, act='relu')
self.pool1 = Pool2D(pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
self.conv2_1 = Conv2D(64, 64, 1, act='relu')
self.conv2_2 = Conv2D(64, 192, 3, padding=1, act='relu')
self.pool2 = Pool2D(pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
self.block3_a = Inception((192, 64), (96, 128), (16, 32), 32)
self.block3_b = Inception((256, 128), (128, 192), (32, 96), 64)
self.pool3 = Pool2D(pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
self.block4_a = Inception((480, 192), (96, 208), (16, 48), 64)
self.block4_b = Inception((512, 160), (112, 224), (24, 64), 64)
self.block4_c = Inception((512, 128), (128, 256), (24, 64), 64)
self.block4_d = Inception((512, 112), (144, 288), (32, 64), 64)
self.block4_e = Inception((528, 256), (160, 320), (32, 128), 128)
self.pool4 = Pool2D(pool_size=3,
pool_stride=2,
pool_padding=1,
pool_type='max')
self.block5_a = Inception((832, 256), (160, 320), (32, 128), 128)
self.block5_b = Inception((832, 384), (192, 384), (48, 128), 128)
self.pool5 = Pool2D(pool_size=7,
pool_stride=1,
global_pooling=True,
pool_type='avg')
self.drop = Dropout(p=0.4)
self.fc = Linear(1024, num_classes)
def __init__(self, num_classes=1):
super(CNN_ConcatConv, self).__init__()
self.conv1 = Conv2D(3, 64, 5, padding=2, stride=1, act='sigmoid')
self.bn1 = BatchNorm(64)
self.conv2 = Conv2D(64, 128, 5, padding=2, stride=1, act='sigmoid')
self.bn2 = BatchNorm(128)
self.conv3 = Conv2D(128, 256, 5, padding=2, stride=1, act='sigmoid')
self.bn3 = BatchNorm(256)
self.conv4 = Conv2D(256, 512, 5, padding=2, stride=1, act='sigmoid')
self.bn4 = BatchNorm(512)
self.block5 = ConcatConv(self.full_name(), (512, 384), (384, 256),
(256, 256), 128)
self.bn5 = BatchNorm(1024)
self.conv6 = Conv2D(1024, 1024, 5, padding=2, stride=1, act='sigmoid')
self.bn6 = BatchNorm(1024)
self.fc1 = Linear(1024 * 7 * 7, 1024, act='sigmoid')
self.fc2 = Linear(1024, num_classes)
self.pool_down = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.dropout = Dropout(p=0.5)
def __init__(self, num_classes=1):
super(CNN_DropOut, self).__init__()
self.conv1 = Conv2D(3, 64, 5, padding=2, stride=1, act='sigmoid')
self.bn1 = BatchNorm(64)
self.conv2 = Conv2D(64, 128, 5, padding=2, stride=1, act='sigmoid')
self.bn2 = BatchNorm(128)
self.conv3 = Conv2D(128, 256, 5, padding=2, stride=1, act='sigmoid')
self.bn3 = BatchNorm(256)
self.conv4 = Conv2D(256, 512, 5, padding=2, stride=1, act='sigmoid')
self.bn4 = BatchNorm(512)
self.conv5 = Conv2D(512, 1024, 5, padding=2, stride=1, act='sigmoid')
self.bn5 = BatchNorm(1024)
self.conv6 = Conv2D(1024, 1024, 5, padding=2, stride=1, act='sigmoid')
self.bn6 = BatchNorm(1024)
self.fc1 = Linear(1024 * 7 * 7, 1024, act='sigmoid')
self.fc2 = Linear(1024, num_classes)
self.pool_down = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
# 按照50%的比例随机丢弃部分神经元
self.dropout = Dropout(p=0.5)
def __init__(self, in_planes):
super(BasicLiner, self).__init__()
self.fc = linear(in_planes, in_planes)
self.bn = norm_layer(in_planes)
self.act = act_layer
self.dp = Dropout()
def __init__(self, num_classes=1):
super(VGG13, self).__init__()
self.drop_out = Dropout(p=0.5)
self.pool = Pool2D(pool_size=3, pool_stride=2, pool_type='max')
self.conv1 = Conv2D(num_channels=3,
num_filters=64,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv2 = Conv2D(num_channels=64,
num_filters=128,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv3 = Conv2D(num_channels=128,
num_filters=256,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv4 = Conv2D(num_channels=256,
num_filters=256,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv5 = Conv2D(num_channels=256,
num_filters=512,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv6 = Conv2D(num_channels=512,
num_filters=512,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv7 = Conv2D(num_channels=512,
num_filters=512,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.conv8 = Conv2D(num_channels=512,
num_filters=512,
filter_size=3,
stride=1,
padding=1,
act='relu')
self.fc1 = Linear(512 * 6 * 6, 4096, act='relu')
self.fc2 = Linear(4096, 4096, act='relu')
self.fc3 = Linear(4096, num_classes, act='relu')
def build_dropout(attr, channels=None, conv_bias=False):
return Dropout(p=attr['dropout_ratio']), channels
def __init__(self,
batch_size,
block_fn,
layers,
num_classes,
data_format='channels_last',
non_local=[],
rep_flow=[],
size=112,
dropout_keep_prob=0.5):
"""Generator for ResNet v1 models.
Args:
block_fn: `function` for the block to use within the model. Either
`residual_block` or `bottleneck_block`.
layers: list of 4 `int`s denoting the number of blocks to include in each
of the 4 block groups. Each group consists of blocks that take inputs of
the same resolution.
num_classes: `int` number of possible classes for image classification.
data_format: `str` either "channels_first" for `[batch, channels, height,
width]` or "channels_last for `[batch, height, width, channels]`.
Returns:
Model `function` that takes in `inputs` and `is_training` and returns the
output `Tensor` of the ResNet model.
"""
super(ResNet3D, self).__init__()
is_training = False
self.stem = Conv2D(
3,
64,
filter_size=7,
bias_attr=fluid.ParamAttr(trainable=False),
stride=2,
padding=3,
param_attr=fluid.initializer.MSRAInitializer(uniform=False))
self.bn1 = BatchNorm(64, act='relu')
self.num_classes = num_classes
self.rep_flow = rf.FlowLayer(batch_size,
block_fn.expansion * 128,
bn=False)
self.flow_conv = Conv2D(512, 512, filter_size=3, stride=1, padding=1)
self.rep_flow2 = rf.FlowLayer(batch_size, 512)
# res 2
inputs = 64
self.res2 = Block3D(block_ind=1,
inputs=inputs,
filters=64,
block_fn=block_fn,
blocks=layers[0],
strides=1,
is_training=is_training,
name='block_group1',
data_format=data_format,
non_local=non_local[0])
# res 3
inputs = 64 * block_fn.expansion
self.res3 = Block3D(block_ind=2,
inputs=inputs,
filters=128,
block_fn=block_fn,
blocks=layers[1],
strides=2,
is_training=is_training,
name='block_group2',
data_format=data_format,
non_local=non_local[1])
# res 4
inputs = 128 * block_fn.expansion
self.res4 = Block3D(block_ind=3,
inputs=inputs,
filters=256,
block_fn=block_fn,
blocks=layers[2],
strides=2,
is_training=is_training,
name='block_group3',
data_format=data_format,
non_local=non_local[2])
# res 5
inputs = 256 * block_fn.expansion
self.res5 = Block3D(block_ind=4,
inputs=inputs,
filters=512,
block_fn=block_fn,
blocks=layers[3],
strides=2,
is_training=is_training,
name='block_group4',
data_format=data_format,
non_local=non_local[3])
self.dropout = Dropout(0.9)
self.classify = Linear(
input_dim=512 * block_fn.expansion,
output_dim=num_classes,
act='softmax',
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.MSRAInitializer(uniform=True)))
