refine sync bn (#4361)

* refine sync bn

* fix bugs of batch_norm

* fix bugs while deploying and modify BatchNorm to BatchNorm2D

* param_attr -> weight_attr in BatchNorm2D

* modify BatchNorm to BatchNorm2D

ppdet/engine/trainer.py

@@ -336,6 +336,12 @@ class Trainer(object):
         assert self.mode == 'train', "Model not in 'train' mode"
         Init_mark = False
 
+        sync_bn = (getattr(self.cfg, 'norm_type', None) == 'sync_bn' and
+                   self.cfg.use_gpu and self._nranks > 1)
+        if sync_bn:
+            self.model = paddle.nn.SyncBatchNorm.convert_sync_batchnorm(
+                self.model)
+
         model = self.model
         if self.cfg.get('fleet', False):
             model = fleet.distributed_model(model)

ppdet/modeling/backbones/blazenet.py

@@ -58,11 +58,8 @@ class ConvBNLayer(nn.Layer):
                 learning_rate=conv_lr, initializer=KaimingNormal()),
             bias_attr=False)
 
-        if norm_type == 'sync_bn':
-            self._batch_norm = nn.SyncBatchNorm(out_channels)
-        else:
-            self._batch_norm = nn.BatchNorm(
-                out_channels, act=None, use_global_stats=False)
+        if norm_type in ['bn', 'sync_bn']:
+            self._batch_norm = nn.BatchNorm2D(out_channels)
 
     def forward(self, x):
         x = self._conv(x)

ppdet/modeling/backbones/hrnet.py

@@ -62,11 +62,11 @@ class ConvNormLayer(nn.Layer):
             learning_rate=norm_lr, regularizer=L2Decay(norm_decay))
         bias_attr = ParamAttr(
             learning_rate=norm_lr, regularizer=L2Decay(norm_decay))
-        global_stats = True if freeze_norm else False
+        global_stats = True if freeze_norm else None
         if norm_type in ['bn', 'sync_bn']:
-            self.norm = nn.BatchNorm(
+            self.norm = nn.BatchNorm2D(
                 ch_out,
-                param_attr=param_attr,
+                weight_attr=param_attr,
                 bias_attr=bias_attr,
                 use_global_stats=global_stats)
         elif norm_type == 'gn':

ppdet/modeling/backbones/lcnet.py

@@ -81,9 +81,9 @@ class ConvBNLayer(nn.Layer):
             weight_attr=ParamAttr(initializer=KaimingNormal()),
             bias_attr=False)
 
-        self.bn = BatchNorm(
+        self.bn = BatchNorm2D(
             num_filters,
-            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
             bias_attr=ParamAttr(regularizer=L2Decay(0.0)))
         self.hardswish = nn.Hardswish()
 

ppdet/modeling/backbones/lite_hrnet.py

@@ -56,11 +56,11 @@ class ConvNormLayer(nn.Layer):
                 regularizer=L2Decay(norm_decay), )
             bias_attr = ParamAttr(
                 learning_rate=norm_lr, regularizer=L2Decay(norm_decay))
-            global_stats = True if freeze_norm else False
+            global_stats = True if freeze_norm else None
             if norm_type in ['bn', 'sync_bn']:
-                self.norm = nn.BatchNorm(
+                self.norm = nn.BatchNorm2D(
                     ch_out,
-                    param_attr=param_attr,
+                    weight_attr=param_attr,
                     bias_attr=bias_attr,
                     use_global_stats=global_stats, )
             elif norm_type == 'gn':
@@ -582,7 +582,7 @@ class LiteHRNetModule(nn.Layer):
                                 stride=1,
                                 padding=0,
                                 bias=False, ),
-                            nn.BatchNorm(self.in_channels[i]),
+                            nn.BatchNorm2D(self.in_channels[i]),
                             nn.Upsample(
                                 scale_factor=2**(j - i), mode='nearest')))
                 elif j == i:
@@ -601,7 +601,7 @@ class LiteHRNetModule(nn.Layer):
                                         padding=1,
                                         groups=self.in_channels[j],
                                         bias=False, ),
-                                    nn.BatchNorm(self.in_channels[j]),
+                                    nn.BatchNorm2D(self.in_channels[j]),
                                     L.Conv2d(
                                         self.in_channels[j],
                                         self.in_channels[i],
@@ -609,7 +609,7 @@ class LiteHRNetModule(nn.Layer):
                                         stride=1,
                                         padding=0,
                                         bias=False, ),
-                                    nn.BatchNorm(self.in_channels[i])))
+                                    nn.BatchNorm2D(self.in_channels[i])))
                         else:
                             conv_downsamples.append(
                                 nn.Sequential(
@@ -621,7 +621,7 @@ class LiteHRNetModule(nn.Layer):
                                         padding=1,
                                         groups=self.in_channels[j],
                                         bias=False, ),
-                                    nn.BatchNorm(self.in_channels[j]),
+                                    nn.BatchNorm2D(self.in_channels[j]),
                                     L.Conv2d(
                                         self.in_channels[j],
                                         self.in_channels[j],
@@ -629,7 +629,7 @@ class LiteHRNetModule(nn.Layer):
                                         stride=1,
                                         padding=0,
                                         bias=False, ),
-                                    nn.BatchNorm(self.in_channels[j]),
+                                    nn.BatchNorm2D(self.in_channels[j]),
                                     nn.ReLU()))
 
                     fuse_layer.append(nn.Sequential(*conv_downsamples))
@@ -777,7 +777,7 @@ class LiteHRNet(nn.Layer):
                                 padding=1,
                                 groups=num_channels_pre_layer[i],
                                 bias=False),
-                            nn.BatchNorm(num_channels_pre_layer[i]),
+                            nn.BatchNorm2D(num_channels_pre_layer[i]),
                             L.Conv2d(
                                 num_channels_pre_layer[i],
                                 num_channels_cur_layer[i],
@@ -785,7 +785,7 @@ class LiteHRNet(nn.Layer):
                                 stride=1,
                                 padding=0,
                                 bias=False, ),
-                            nn.BatchNorm(num_channels_cur_layer[i]),
+                            nn.BatchNorm2D(num_channels_cur_layer[i]),
                             nn.ReLU()))
                 else:
                     transition_layers.append(None)
@@ -802,7 +802,7 @@ class LiteHRNet(nn.Layer):
                                 stride=2,
                                 padding=1,
                                 bias=False, ),
-                            nn.BatchNorm(num_channels_pre_layer[-1]),
+                            nn.BatchNorm2D(num_channels_pre_layer[-1]),
                             L.Conv2d(
                                 num_channels_pre_layer[-1],
                                 num_channels_cur_layer[i]
@@ -812,7 +812,7 @@ class LiteHRNet(nn.Layer):
                                 stride=1,
                                 padding=0,
                                 bias=False, ),
-                            nn.BatchNorm(num_channels_cur_layer[i]
+                            nn.BatchNorm2D(num_channels_cur_layer[i]
                                            if j == i - num_branches_pre else
                                            num_channels_pre_layer[-1]),
                             nn.ReLU()))

ppdet/modeling/backbones/mobilenet_v1.py

@@ -59,16 +59,9 @@ class ConvBNLayer(nn.Layer):
 
         param_attr = ParamAttr(regularizer=L2Decay(norm_decay))
         bias_attr = ParamAttr(regularizer=L2Decay(norm_decay))
-        if norm_type == 'sync_bn':
-            self._batch_norm = nn.SyncBatchNorm(
+        if norm_type in ['sync_bn', 'bn']:
+            self._batch_norm = nn.BatchNorm2D(
                 out_channels, weight_attr=param_attr, bias_attr=bias_attr)
-        else:
-            self._batch_norm = nn.BatchNorm(
-                out_channels,
-                act=None,
-                param_attr=param_attr,
-                bias_attr=bias_attr,
-                use_global_stats=False)
 
     def forward(self, x):
         x = self._conv(x)

ppdet/modeling/backbones/mobilenet_v3.py

@@ -74,15 +74,11 @@ class ConvBNLayer(nn.Layer):
             learning_rate=norm_lr,
             regularizer=L2Decay(norm_decay),
             trainable=False if freeze_norm else True)
-        global_stats = True if freeze_norm else False
-        if norm_type == 'sync_bn':
-            self.bn = nn.SyncBatchNorm(
-                out_c, weight_attr=param_attr, bias_attr=bias_attr)
-        else:
-            self.bn = nn.BatchNorm(
+        global_stats = True if freeze_norm else None
+        if norm_type in ['sync_bn', 'bn']:
+            self.bn = nn.BatchNorm2D(
                 out_c,
-                act=None,
-                param_attr=param_attr,
+                weight_attr=param_attr,
                 bias_attr=bias_attr,
                 use_global_stats=global_stats)
         norm_params = self.bn.parameters()

ppdet/modeling/backbones/resnet.py

@@ -100,15 +100,11 @@ class ConvNormLayer(nn.Layer):
             regularizer=L2Decay(norm_decay),
             trainable=False if freeze_norm else True)
 
-        global_stats = True if freeze_norm else False
-        if norm_type == 'sync_bn':
-            self.norm = nn.SyncBatchNorm(
-                ch_out, weight_attr=param_attr, bias_attr=bias_attr)
-        else:
-            self.norm = nn.BatchNorm(
+        global_stats = True if freeze_norm else None
+        if norm_type in ['sync_bn', 'bn']:
+            self.norm = nn.BatchNorm2D(
                 ch_out,
-                act=None,
-                param_attr=param_attr,
+                weight_attr=param_attr,
                 bias_attr=bias_attr,
                 use_global_stats=global_stats)
         norm_params = self.norm.parameters()

ppdet/modeling/backbones/shufflenet_v2.py

@@ -51,15 +51,17 @@ class ConvBNLayer(nn.Layer):
             weight_attr=ParamAttr(initializer=KaimingNormal()),
             bias_attr=False)
 
-        self._batch_norm = BatchNorm(
+        self._batch_norm = BatchNorm2D(
             out_channels,
-            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
-            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
-            act=act)
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)))
+        self.act = act
 
     def forward(self, inputs):
         y = self._conv(inputs)
         y = self._batch_norm(y)
+        if self.act:
+            y = getattr(F, self.act)(y)
         return y
 
 

ppdet/modeling/layers.py

@@ -174,12 +174,9 @@ class ConvNormLayer(nn.Layer):
         bias_attr = ParamAttr(
             learning_rate=norm_lr,
             regularizer=L2Decay(norm_decay) if norm_decay is not None else None)
-        if norm_type == 'bn':
+        if norm_type in ['bn', 'sync_bn']:
             self.norm = nn.BatchNorm2D(
                 ch_out, weight_attr=param_attr, bias_attr=bias_attr)
-        elif norm_type == 'sync_bn':
-            self.norm = nn.SyncBatchNorm(
-                ch_out, weight_attr=param_attr, bias_attr=bias_attr)
         elif norm_type == 'gn':
             self.norm = nn.GroupNorm(
                 num_groups=norm_groups,

ppdet/modeling/necks/bifpn.py

@@ -52,10 +52,8 @@ class SeparableConvLayer(nn.Layer):
         self.pointwise_conv = nn.Conv2D(in_channels, self.out_channels, 1)
 
         # norm type
-        if self.norm_type == 'bn':
+        if self.norm_type in ['bn', 'sync_bn']:
             self.norm = nn.BatchNorm2D(self.out_channels)
-        elif self.norm_type == 'sync_bn':
-            self.norm = nn.SyncBatchNorm(self.out_channels)
         elif self.norm_type == 'gn':
             self.norm = nn.GroupNorm(
                 num_groups=self.norm_groups, num_channels=self.out_channels)

ppdet/modeling/necks/blazeface_fpn.py

@@ -54,11 +54,8 @@ class ConvBNLayer(nn.Layer):
                 learning_rate=conv_lr, initializer=KaimingNormal()),
             bias_attr=False)
 
-        if norm_type == 'sync_bn':
-            self._batch_norm = nn.SyncBatchNorm(out_channels)
-        else:
-            self._batch_norm = nn.BatchNorm(
-                out_channels, act=None, use_global_stats=False)
+        if norm_type in ['sync_bn', 'bn']:
+            self._batch_norm = nn.BatchNorm2D(out_channels)
 
     def forward(self, x):
         x = self._conv(x)

ppdet/modeling/ops.py

@@ -50,10 +50,6 @@ def batch_norm(ch,
                freeze_norm=False,
                initializer=None,
                data_format='NCHW'):
-    if norm_type == 'sync_bn':
-        batch_norm = nn.SyncBatchNorm
-    else:
-        batch_norm = nn.BatchNorm2D
 
     norm_lr = 0. if freeze_norm else 1.
     weight_attr = ParamAttr(
@@ -66,7 +62,8 @@ def batch_norm(ch,
         regularizer=L2Decay(norm_decay),
         trainable=False if freeze_norm else True)
 
-    norm_layer = batch_norm(
+    if norm_type in ['sync_bn', 'bn']:
+        norm_layer = nn.BatchNorm2D(
             ch,
             weight_attr=weight_attr,
             bias_attr=bias_attr,

ppdet/modeling/reid/pplcnet_embedding.py

@@ -21,7 +21,7 @@ import paddle.nn as nn
 import paddle.nn.functional as F
 from paddle.nn.initializer import Normal, Constant
 from paddle import ParamAttr
-from paddle.nn import AdaptiveAvgPool2D, BatchNorm, Conv2D, Linear
+from paddle.nn import AdaptiveAvgPool2D, BatchNorm2D, Conv2D, Linear
 from paddle.regularizer import L2Decay
 from paddle.nn.initializer import KaimingNormal, XavierNormal
 from ppdet.core.workspace import register
@@ -77,9 +77,9 @@ class ConvBNLayer(nn.Layer):
             weight_attr=ParamAttr(initializer=KaimingNormal()),
             bias_attr=False)
 
-        self.bn = BatchNorm(
+        self.bn = BatchNorm2D(
             num_filters,
-            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
             bias_attr=ParamAttr(regularizer=L2Decay(0.0)))
         self.hardswish = nn.Hardswish()
 

ppdet/modeling/reid/resnet.py

@@ -55,12 +55,14 @@ class ConvBNLayer(nn.Layer):
             bias_attr=False,
             data_format=data_format)
 
-        self._batch_norm = nn.BatchNorm(
-            num_filters, act=act, data_layout=data_format)
+        self._batch_norm = nn.BatchNorm2D(num_filters, data_layout=data_format)
+        self.act = act
 
     def forward(self, inputs):
         y = self._conv(inputs)
         y = self._batch_norm(y)
+        if self.act:
+            y = getattr(F, self.act)(y)
         return y