Use common layers

paddlers/rs_models/cd/bit.py

@@ -18,7 +18,7 @@ import paddle.nn.functional as F
 from paddle.nn.initializer import Normal
 
 from .backbones import resnet
-from .layers import Conv3x3, Conv1x1, get_norm_layer, Identity
+from .layers import Conv3x3, Conv1x1, get_bn_layer, Identity
 from .param_init import KaimingInitMixin
 
 
@@ -369,12 +369,12 @@ class Backbone(nn.Layer, KaimingInitMixin):
             self.resnet = resnet.resnet18(
                 pretrained=pretrained,
                 strides=strides,
-                norm_layer=get_norm_layer())
+                norm_layer=get_bn_layer())
         elif arch == 'resnet34':
             self.resnet = resnet.resnet34(
                 pretrained=pretrained,
                 strides=strides,
-                norm_layer=get_norm_layer())
+                norm_layer=get_bn_layer())
         else:
             raise ValueError
 

paddlers/rs_models/cd/cdnet.py

@@ -15,10 +15,25 @@
 import paddle
 import paddle.nn as nn
 
+from .layers import Conv7x7
+
 
 class CDNet(nn.Layer):
-    def __init__(self, in_channels=6, num_classes=2):
+    """
+    The CDNet implementation based on PaddlePaddle.
+
+    The original article refers to
+        Pablo F. Alcantarilla, et al., "Street-View Change Detection with Deconvolutional Networks"
+        (https://link.springer.com/article/10.1007/s10514-018-9734-5).
+
+    Args:
+        in_channels (int): The number of bands of the input images.
+        num_classes (int): The number of target classes.
+    """
+
+    def __init__(self, in_channels, num_classes):
         super(CDNet, self).__init__()
+
         self.conv1 = Conv7x7(in_channels, 64, norm=True, act=True)
         self.pool1 = nn.MaxPool2D(2, 2, return_mask=True)
         self.conv2 = Conv7x7(64, 64, norm=True, act=True)
@@ -28,6 +43,7 @@ class CDNet(nn.Layer):
         self.conv4 = Conv7x7(64, 64, norm=True, act=True)
         self.pool4 = nn.MaxPool2D(2, 2, return_mask=True)
         self.conv5 = Conv7x7(64, 64, norm=True, act=True)
+
         self.upool4 = nn.MaxUnPool2D(2, 2)
         self.conv6 = Conv7x7(64, 64, norm=True, act=True)
         self.upool3 = nn.MaxUnPool2D(2, 2)
@@ -39,37 +55,15 @@ class CDNet(nn.Layer):
 
     def forward(self, t1, t2):
         x = paddle.concat([t1, t2], axis=1)
+
         x, ind1 = self.pool1(self.conv1(x))
         x, ind2 = self.pool2(self.conv2(x))
         x, ind3 = self.pool3(self.conv3(x))
         x, ind4 = self.pool4(self.conv4(x))
+
         x = self.conv5(self.upool4(x, ind4))
         x = self.conv6(self.upool3(x, ind3))
         x = self.conv7(self.upool2(x, ind2))
         x = self.conv8(self.upool1(x, ind1))
-        return [self.conv_out(x)]
-
-
-class Conv7x7(nn.Layer):
-    def __init__(self, in_ch, out_ch, norm=False, act=False):
-        super(Conv7x7, self).__init__()
-        layers = [
-            nn.Pad2D(3), nn.Conv2D(
-                in_ch, out_ch, 7, bias_attr=(False if norm else None))
-        ]
-        if norm:
-            layers.append(nn.BatchNorm2D(out_ch))
-        if act:
-            layers.append(nn.ReLU())
-        self.layers = nn.Sequential(*layers)
 
-    def forward(self, x):
-        return self.layers(x)
-
-
-if __name__ == "__main__":
-    t1 = paddle.randn((1, 3, 512, 512), dtype="float32")
-    t2 = paddle.randn((1, 3, 512, 512), dtype="float32")
-    model = CDNet(6, 2)
-    pred = model(t1, t2)[0]
-    print(pred.shape)
+        return [self.conv_out(x)]

paddlers/rs_models/cd/dsamnet.py

@@ -16,7 +16,7 @@ import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
 
-from .layers import make_norm, Conv3x3, CBAM
+from .layers import make_bn, Conv3x3, CBAM
 from .stanet import Backbone, Decoder
 
 
@@ -93,7 +93,7 @@ class DSLayer(nn.Sequential):
         super(DSLayer, self).__init__(
             nn.Conv2DTranspose(
                 in_ch, itm_ch, kernel_size=3, padding=1, **convd_kwargs),
-            make_norm(itm_ch),
+            make_bn(itm_ch),
             nn.ReLU(),
             nn.Dropout2D(p=0.2),
             nn.Conv2DTranspose(

paddlers/rs_models/cd/dsifn.py

@@ -20,7 +20,7 @@ import paddle.nn as nn
 import paddle.nn.functional as F
 from paddle.vision.models import vgg16
 
-from .layers import Conv1x1, make_norm, ChannelAttention, SpatialAttention
+from .layers import Conv1x1, make_bn, ChannelAttention, SpatialAttention
 
 
 class DSIFN(nn.Layer):
@@ -52,19 +52,19 @@ class DSIFN(nn.Layer):
         self.sa5 = SpatialAttention()
 
         self.ca1 = ChannelAttention(in_ch=1024)
-        self.bn_ca1 = make_norm(1024)
+        self.bn_ca1 = make_bn(1024)
         self.o1_conv1 = conv2d_bn(1024, 512, use_dropout)
         self.o1_conv2 = conv2d_bn(512, 512, use_dropout)
-        self.bn_sa1 = make_norm(512)
+        self.bn_sa1 = make_bn(512)
         self.o1_conv3 = Conv1x1(512, num_classes)
         self.trans_conv1 = nn.Conv2DTranspose(512, 512, kernel_size=2, stride=2)
 
         self.ca2 = ChannelAttention(in_ch=1536)
-        self.bn_ca2 = make_norm(1536)
+        self.bn_ca2 = make_bn(1536)
         self.o2_conv1 = conv2d_bn(1536, 512, use_dropout)
         self.o2_conv2 = conv2d_bn(512, 256, use_dropout)
         self.o2_conv3 = conv2d_bn(256, 256, use_dropout)
-        self.bn_sa2 = make_norm(256)
+        self.bn_sa2 = make_bn(256)
         self.o2_conv4 = Conv1x1(256, num_classes)
         self.trans_conv2 = nn.Conv2DTranspose(256, 256, kernel_size=2, stride=2)
 
@@ -72,7 +72,7 @@ class DSIFN(nn.Layer):
         self.o3_conv1 = conv2d_bn(768, 256, use_dropout)
         self.o3_conv2 = conv2d_bn(256, 128, use_dropout)
         self.o3_conv3 = conv2d_bn(128, 128, use_dropout)
-        self.bn_sa3 = make_norm(128)
+        self.bn_sa3 = make_bn(128)
         self.o3_conv4 = Conv1x1(128, num_classes)
         self.trans_conv3 = nn.Conv2DTranspose(128, 128, kernel_size=2, stride=2)
 
@@ -80,7 +80,7 @@ class DSIFN(nn.Layer):
         self.o4_conv1 = conv2d_bn(384, 128, use_dropout)
         self.o4_conv2 = conv2d_bn(128, 64, use_dropout)
         self.o4_conv3 = conv2d_bn(64, 64, use_dropout)
-        self.bn_sa4 = make_norm(64)
+        self.bn_sa4 = make_bn(64)
         self.o4_conv4 = Conv1x1(64, num_classes)
         self.trans_conv4 = nn.Conv2DTranspose(64, 64, kernel_size=2, stride=2)
 
@@ -88,7 +88,7 @@ class DSIFN(nn.Layer):
         self.o5_conv1 = conv2d_bn(192, 64, use_dropout)
         self.o5_conv2 = conv2d_bn(64, 32, use_dropout)
         self.o5_conv3 = conv2d_bn(32, 16, use_dropout)
-        self.bn_sa5 = make_norm(16)
+        self.bn_sa5 = make_bn(16)
         self.o5_conv4 = Conv1x1(16, num_classes)
 
         self.init_weight()
@@ -211,7 +211,7 @@ def conv2d_bn(in_ch, out_ch, with_dropout=True):
         nn.Conv2D(
             in_ch, out_ch, kernel_size=3, stride=1, padding=1),
         nn.PReLU(),
-        make_norm(out_ch),
+        make_bn(out_ch),
     ]
     if with_dropout:
         lst.append(nn.Dropout(p=0.6))

paddlers/rs_models/cd/fc_ef.py

@@ -26,13 +26,13 @@ class FCEarlyFusion(nn.Layer):
     The FC-EF implementation based on PaddlePaddle.
 
     The original article refers to
-        Caye Daudt, R., et al. "Fully convolutional siamese networks for change detection"
+        Rodrigo Caye Daudt, et al. "Fully convolutional siamese networks for change detection"
         (https://arxiv.org/abs/1810.08462).
 
     Args:
         in_channels (int): The number of bands of the input images.
         num_classes (int): The number of target classes.
-        use_dropout (bool, optional): A bool value that indicates whether to use dropout layers. When the model is trained 
+        use_dropout (bool, optional): A bool value that indicates whether to use dropout layers. When the model is trained
             on a relatively small dataset, the dropout layers help prevent overfitting. Default: False.
     """
 

paddlers/rs_models/cd/fc_siam_conc.py

@@ -26,7 +26,7 @@ class FCSiamConc(nn.Layer):
     The FC-Siam-conc implementation based on PaddlePaddle.
 
     The original article refers to
-        Caye Daudt, R., et al. "Fully convolutional siamese networks for change detection"
+        Rodrigo Caye Daudt, et al. "Fully convolutional siamese networks for change detection"
         (https://arxiv.org/abs/1810.08462).
 
     Args:

paddlers/rs_models/cd/fc_siam_diff.py

@@ -26,7 +26,7 @@ class FCSiamDiff(nn.Layer):
     The FC-Siam-diff implementation based on PaddlePaddle.
 
     The original article refers to
-        Caye Daudt, R., et al. "Fully convolutional siamese networks for change detection"
+        Rodrigo Caye Daudt, et al. "Fully convolutional siamese networks for change detection"
         (https://arxiv.org/abs/1810.08462).
 
     Args:

paddlers/rs_models/cd/layers/__init__.py

@@ -12,5 +12,5 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .blocks import *
+from ...layers.blocks import *
 from .attention import ChannelAttention, SpatialAttention, CBAM

paddlers/rs_models/cd/snunet.py

@@ -18,7 +18,7 @@ import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
 
-from .layers import Conv1x1, MaxPool2x2, make_norm, ChannelAttention
+from .layers import Conv1x1, MaxPool2x2, make_bn, ChannelAttention
 from .param_init import KaimingInitMixin
 
 
@@ -145,9 +145,9 @@ class ConvBlockNested(nn.Layer):
         super(ConvBlockNested, self).__init__()
         self.act = nn.ReLU()
         self.conv1 = nn.Conv2D(in_ch, mid_ch, kernel_size=3, padding=1)
-        self.bn1 = make_norm(mid_ch)
+        self.bn1 = make_bn(mid_ch)
         self.conv2 = nn.Conv2D(mid_ch, out_ch, kernel_size=3, padding=1)
-        self.bn2 = make_norm(out_ch)
+        self.bn2 = make_bn(out_ch)
 
     def forward(self, x):
         x = self.conv1(x)

paddlers/rs_models/cd/stanet.py

@@ -17,7 +17,7 @@ import paddle.nn as nn
 import paddle.nn.functional as F
 
 from .backbones import resnet
-from .layers import Conv1x1, Conv3x3, get_norm_layer, Identity
+from .layers import Conv1x1, Conv3x3, get_bn_layer, Identity
 from .param_init import KaimingInitMixin
 
 
@@ -100,17 +100,17 @@ class Backbone(nn.Layer, KaimingInitMixin):
             self.resnet = resnet.resnet18(
                 pretrained=pretrained,
                 strides=strides,
-                norm_layer=get_norm_layer())
+                norm_layer=get_bn_layer())
         elif arch == 'resnet34':
             self.resnet = resnet.resnet34(
                 pretrained=pretrained,
                 strides=strides,
-                norm_layer=get_norm_layer())
+                norm_layer=get_bn_layer())
         elif arch == 'resnet50':
             self.resnet = resnet.resnet50(
                 pretrained=pretrained,
                 strides=strides,
-                norm_layer=get_norm_layer())
+                norm_layer=get_bn_layer())
         else:
             raise ValueError
 

paddlers/rs_models/clas/condensenet_v2.py

@@ -20,6 +20,8 @@ Apache License [see LICENSE for details]
 import paddle
 import paddle.nn as nn
 
+from ...layers.blocks import make_bn
+
 __all__ = ["CondenseNetV2_a", "CondenseNetV2_b", "CondenseNetV2_c"]
 
 
@@ -63,9 +65,7 @@ class Conv(nn.Sequential):
             activation="ReLU",
             bn_momentum=0.9, ):
         super(Conv, self).__init__()
-        self.add_sublayer(
-            "norm", nn.BatchNorm2D(
-                in_channels, momentum=bn_momentum))
+        self.add_sublayer("norm", make_bn(in_channels, momentum=bn_momentum))
         if activation == "ReLU":
             self.add_sublayer("activation", nn.ReLU())
         elif activation == "HS":
@@ -122,7 +122,7 @@ class CondenseLGC(nn.Layer):
         self.in_channels = in_channels
         self.out_channels = out_channels
         self.groups = groups
-        self.norm = nn.BatchNorm2D(self.in_channels)
+        self.norm = make_bn(self.in_channels)
         if activation == "ReLU":
             self.activation = nn.ReLU()
         elif activation == "HS":
@@ -164,7 +164,7 @@ class CondenseSFR(nn.Layer):
         self.in_channels = in_channels
         self.out_channels = out_channels
         self.groups = groups
-        self.norm = nn.BatchNorm2D(self.in_channels)
+        self.norm = make_bn(self.in_channels)
         if activation == "ReLU":
             self.activation = nn.ReLU()
         elif activation == "HS":
@@ -361,8 +361,7 @@ class CondenseNetV2(nn.Layer):
             trans = _Transition()
             self.features.add_sublayer("transition_%d" % (i + 1), trans)
         else:
-            self.features.add_sublayer("norm_last",
-                                       nn.BatchNorm2D(self.num_features))
+            self.features.add_sublayer("norm_last", make_bn(self.num_features))
             self.features.add_sublayer("relu_last", nn.ReLU())
             self.features.add_sublayer("pool_last",
                                        nn.AvgPool2D(self.pool_size))
@@ -389,7 +388,7 @@ class CondenseNetV2(nn.Layer):
         for m in self.sublayers():
             if isinstance(m, nn.Conv2D):
                 nn.initializer.KaimingNormal()(m.weight)
-            elif isinstance(m, nn.BatchNorm2D):
+            elif isinstance(m, (nn.BatchNorm2D, nn.SyncBatchNorm)):
                 nn.initializer.Constant(value=1.0)(m.weight)
                 nn.initializer.Constant(value=0.0)(m.bias)
 

paddlers/rs_models/layers/__init__.py

@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

paddlers/rs_models/layers/blocks.py

@@ -12,18 +12,25 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
+
+import paddle
 import paddle.nn as nn
 
 __all__ = [
     'BasicConv', 'Conv1x1', 'Conv3x3', 'Conv7x7', 'MaxPool2x2', 'MaxUnPool2x2',
-    'ConvTransposed3x3', 'Identity', 'get_norm_layer', 'get_act_layer',
-    'make_norm', 'make_act'
+    'ConvTransposed3x3', 'Identity', 'get_bn_layer', 'get_act_layer', 'make_bn',
+    'make_act'
 ]
 
 
-def get_norm_layer():
-    # TODO: select appropriate norm layer.
-    return nn.BatchNorm2D
+def get_bn_layer():
+    if paddle.get_device() == 'cpu' or os.environ.get('PADDLESEG_EXPORT_STAGE'):
+        return nn.BatchNorm2D
+    elif paddle.distributed.ParallelEnv().nranks == 1:
+        return nn.BatchNorm2D
+    else:
+        return nn.SyncBatchNorm
 
 
 def get_act_layer():
@@ -31,8 +38,8 @@ def get_act_layer():
     return nn.ReLU
 
 
-def make_norm(*args, **kwargs):
-    norm_layer = get_norm_layer()
+def make_bn(*args, **kwargs):
+    norm_layer = get_bn_layer()
     return norm_layer(*args, **kwargs)
 
 
@@ -66,7 +73,7 @@ class BasicConv(nn.Layer):
                 **kwargs))
         if norm:
             if norm is True:
-                norm = make_norm(out_ch)
+                norm = make_bn(out_ch)
             seq.append(norm)
         if act:
             if act is True:
@@ -171,7 +178,7 @@ class ConvTransposed3x3(nn.Layer):
                 **kwargs))
         if norm:
             if norm is True:
-                norm = make_norm(out_ch)
+                norm = make_bn(out_ch)
             seq.append(norm)
         if act:
             if act is True:

paddlers/rs_models/seg/farseg.py

@@ -25,7 +25,8 @@ from paddle.vision.models import resnet50
 from paddle import nn
 import paddle.nn.functional as F
 
-from .layers import (Identity, ConvReLU, kaiming_normal_init, constant_init)
+from .layers import (Identity, Conv3x3, Conv1x1, get_bn_layer,
+                     kaiming_normal_init, constant_init)
 
 
 class FPN(nn.Layer):
@@ -35,11 +36,7 @@ class FPN(nn.Layer):
         order, and must be consecutive
     """
 
-    def __init__(self,
-                 in_channels_list,
-                 out_channels,
-                 conv_block=ConvReLU,
-                 top_blocks=None):
+    def __init__(self, in_channels_list, out_channels, top_blocks=None):
         super(FPN, self).__init__()
 
         inner_blocks = []
@@ -47,8 +44,10 @@ class FPN(nn.Layer):
         for idx, in_channels in enumerate(in_channels_list, 1):
             if in_channels == 0:
                 continue
-            inner_block_module = conv_block(in_channels, out_channels, 1)
-            layer_block_module = conv_block(out_channels, out_channels, 3, 1)
+            inner_block_module = Conv1x1(
+                in_channels, out_channels, norm=False, act=True)
+            layer_block_module = Conv3x3(
+                out_channels, out_channels, norm=False, act=True)
             for module in [inner_block_module, layer_block_module]:
                 for m in module.sublayers():
                     if isinstance(m, nn.Conv2D):
@@ -131,13 +130,11 @@ class SceneRelation(nn.Layer):
         self.feature_reencoders = nn.LayerList()
         for c in channel_list:
             self.content_encoders.append(
-                nn.Sequential(
-                    nn.Conv2D(c, out_channels, 1),
-                    nn.BatchNorm2D(out_channels), nn.ReLU()))
+                Conv1x1(
+                    c, out_channels, norm=True, act=True))
             self.feature_reencoders.append(
-                nn.Sequential(
-                    nn.Conv2D(c, out_channels, 1),
-                    nn.BatchNorm2D(out_channels), nn.ReLU()))
+                Conv1x1(
+                    c, out_channels, norm=True, act=True))
         self.normalizer = nn.Sigmoid()
 
     def forward(self, scene_feature, features: list):
@@ -170,19 +167,22 @@ class AssymetricDecoder(nn.Layer):
                  out_channels,
                  in_feat_output_strides=(4, 8, 16, 32),
                  out_feat_output_stride=4,
-                 norm_fn=nn.BatchNorm2D,
+                 norm_fn='batch_norm',
                  num_groups_gn=None):
         super(AssymetricDecoder, self).__init__()
-        if norm_fn == nn.BatchNorm2D:
+        if norm_fn == 'batch_norm':
+            norm_fn = get_bn_layer()
             norm_fn_args = dict(num_features=out_channels)
-        elif norm_fn == nn.GroupNorm:
+        elif norm_fn == 'group_norm':
+            norm_fn = nn.GroupNorm
             if num_groups_gn is None:
                 raise ValueError(
-                    'When norm_fn is nn.GroupNorm, num_groups_gn is needed.')
+                    'When norm_fn is group_norm, `num_groups_gn` is needed.')
             norm_fn_args = dict(
                 num_groups=num_groups_gn, num_channels=out_channels)
         else:
-            raise ValueError('Type of {} is not support.'.format(type(norm_fn)))
+            raise ValueError('{} is not a supported normalization type.'.format(
+                norm_fn))
         self.blocks = nn.LayerList()
         for in_feat_os in in_feat_output_strides:
             num_upsample = int(math.log2(int(in_feat_os))) - int(

paddlers/rs_models/seg/layers/layers_lib.py

@@ -16,72 +16,7 @@ import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
 
-
-class ConvBNReLU(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 out_channels,
-                 kernel_size,
-                 padding='same',
-                 **kwargs):
-        super().__init__()
-        self._conv = nn.Conv2D(
-            in_channels, out_channels, kernel_size, padding=padding, **kwargs)
-        if 'data_format' in kwargs:
-            data_format = kwargs['data_format']
-        else:
-            data_format = 'NCHW'
-        self._batch_norm = nn.BatchNorm2D(out_channels, data_format=data_format)
-
-    def forward(self, x):
-        x = self._conv(x)
-        x = self._batch_norm(x)
-        x = F.relu(x)
-        return x
-
-
-class ConvBN(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 out_channels,
-                 kernel_size,
-                 padding='same',
-                 **kwargs):
-        super().__init__()
-        self._conv = nn.Conv2D(
-            in_channels, out_channels, kernel_size, padding=padding, **kwargs)
-        if 'data_format' in kwargs:
-            data_format = kwargs['data_format']
-        else:
-            data_format = 'NCHW'
-        self._batch_norm = nn.BatchNorm2D(out_channels, data_format=data_format)
-
-    def forward(self, x):
-        x = self._conv(x)
-        x = self._batch_norm(x)
-        return x
-
-
-class ConvReLU(nn.Layer):
-    def __init__(self,
-                 in_channels,
-                 out_channels,
-                 kernel_size,
-                 padding='same',
-                 **kwargs):
-        super().__init__()
-        self._conv = nn.Conv2D(
-            in_channels, out_channels, kernel_size, padding=padding, **kwargs)
-        if 'data_format' in kwargs:
-            data_format = kwargs['data_format']
-        else:
-            data_format = 'NCHW'
-        self._relu = nn.ReLU()
-
-    def forward(self, x):
-        x = self._conv(x)
-        x = self._relu(x)
-        return x
+from ...layers.blocks import *
 
 
 class Add(nn.Layer):
@@ -95,15 +30,19 @@ class Add(nn.Layer):
 class Activation(nn.Layer):
     """
     The wrapper of activations.
+
     Args:
         act (str, optional): The activation name in lowercase. It must be one of ['elu', 'gelu',
             'hardshrink', 'tanh', 'hardtanh', 'prelu', 'relu', 'relu6', 'selu', 'leakyrelu', 'sigmoid',
             'softmax', 'softplus', 'softshrink', 'softsign', 'tanhshrink', 'logsigmoid', 'logsoftmax',
             'hsigmoid']. Default: None, means identical transformation.
+
     Returns:
         A callable object of Activation.
+
     Raises:
         KeyError: When parameter `act` is not in the optional range.
+        
     Examples:
         from paddleseg.models.common.activation import Activation
         relu = Activation("relu")
@@ -138,11 +77,3 @@ class Activation(nn.Layer):
             return self.act_func(x)
         else:
             return x
-
-
-class Identity(nn.Layer):
-    def __init__(self, *args, **kwargs):
-        super(Identity, self).__init__()
-
-    def forward(self, input):
-        return input