add FEPAN and head

ppocr/modeling/heads/det_db_head.py

@@ -31,13 +31,14 @@ def get_bias_attr(k):
 
 
 class Head(nn.Layer):
-    def __init__(self, in_channels, name_list):
+    def __init__(self, in_channels, name_list, kernel_list=[3, 2, 2]):
         super(Head, self).__init__()
+
         self.conv1 = nn.Conv2D(
             in_channels=in_channels,
             out_channels=in_channels // 4,
-            kernel_size=3,
+            kernel_size=kernel_size[0],
-            padding=1,
+            padding=int(kernel_size[0] // 2),
             weight_attr=ParamAttr(),
             bias_attr=False)
         self.conv_bn1 = nn.BatchNorm(
@@ -50,7 +51,7 @@ class Head(nn.Layer):
         self.conv2 = nn.Conv2DTranspose(
             in_channels=in_channels // 4,
             out_channels=in_channels // 4,
-            kernel_size=2,
+            kernel_size=kernel_size[1],
             stride=2,
             weight_attr=ParamAttr(
                 initializer=paddle.nn.initializer.KaimingUniform()),
@@ -65,7 +66,7 @@ class Head(nn.Layer):
         self.conv3 = nn.Conv2DTranspose(
             in_channels=in_channels // 4,
             out_channels=1,
-            kernel_size=2,
+            kernel_size=kernel_size[1],
             stride=2,
             weight_attr=ParamAttr(
                 initializer=paddle.nn.initializer.KaimingUniform()),

ppocr/modeling/necks/db_fpn.py

@@ -20,7 +20,7 @@ import paddle
 from paddle import nn
 import paddle.nn.functional as F
 from paddle import ParamAttr
-from ppocr.backbones.det_mobilenet_v3 import SEModule
+from ppocr.backbones.det_mobilenet_v3 import SEModule, ConvBNLayer
 
 
 class DBFPN(nn.Layer):
@@ -179,3 +179,85 @@ class CAFPN(nn.Layer):
 
         fuse = paddle.concat([p5, p4, p3, p2], axis=1)
         return fuse
+
+
+class FEPAN(nn.Layer):
+    def __init__(self, in_channels, out_channels, **kwargs):
+        super(FEPAN, self).__init__()
+        self.out_channels = out_channels
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+
+        self.ins_convs = []
+        self.inp_convs = []
+        # pan head
+        self.pan_head_conv = []
+        self.pan_lat_conv = []
+
+        for i in range(len(in_channels)):
+            self.ins_conv.append(
+                nn.Conv2D(
+                    in_channels=in_channels[0],
+                    out_channels=self.out_channels,
+                    kernel_size=1,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False))
+
+            self.inp_conv.append(
+                ConvBNLayer(
+                    in_channels=self.out_channels,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                    padding=4))
+
+            if i > 0:
+                self.pan_head_conv.append(
+                    nn.Conv2D(
+                        in_channels=self.out_channels // 4,
+                        out_channels=self.out_channels // 4,
+                        kernel_size=3,
+                        padding=1,
+                        stride=2,
+                        weight_attr=ParamAttr(initializer=weight_attr),
+                        bias_attr=False))
+            self.pan_lat_conv.append(
+                ConvBNLayer(
+                    in_channels=self.out_channels // 4,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                    padding=4))
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.ins_conv[3](c5)
+        in4 = self.ins_conv[2](c4)
+        in3 = self.ins_conv[1](c3)
+        in2 = self.ins_conv[0](c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1)  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1)  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1)  # 1/4
+
+        f5 = self.inp_conv[3](in5)
+        f4 = self.inp_conv[2](out4)
+        f3 = self.inp_conv[1](out3)
+        f2 = self.inp_conv[0](out2)
+
+        pan3 = f3 + self.pan_head[0](f2)
+        pan4 = f4 + self.pan_head[1](pan3)
+        pan5 = f5 + self.pan_head[2](pan4)
+
+        p2 = self.pan_lat[0](f2)
+        p3 = self.pan_lat[1](pan3)
+        p4 = self.pan_lat[2](pan4)
+        p5 = self.pan_lat[3](pan5)
+
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+        return fuse