added skeleton

efficient_net/config.py

@@ -0,0 +1,16 @@
+#!/usr/bin/env python
+
+
+class MBConfig(object):
+
+    IN_CHANNELS = None
+
+    OUT_CHANNELS = None
+
+    EXPANSION_RATIO = 1
+
+
+class ENConfig(object):
+
+    # MBConv config
+

efficient_net/network.py

@@ -0,0 +1,77 @@
+#!/usr/bin/env python
+
+import torch
+import torch.nn as nn
+
+
+class ConvBNR(nn.Module):
+
+    def __init__(self,
+                 use_bn: bool=True,
+                 use_relu: bool=True,
+                 **kwargs: dict):
+        super(ConvBNR, self).__init__()
+        self.bn = None
+        self.relu = None
+
+        if use_bn:
+            self.bn = nn.BatchNorm2d(num_features=kwargs['out_channels'],
+                                     momentum=kwargs.pop('bn_momentum'),
+                                     eps=kwargs.pop('bn_eps'))
+        self.conv = nn.Conv2d(**kwargs)
+        if use_relu:
+            self.relu = nn.ReLU()
+
+    def forward(self, inp):
+        x = self.conv(inp)
+        if self.bn is not None:
+            x = self.bn(x)
+        if self.relu is not None:
+            x = self.relu(x)
+        return x
+
+
+class MBConvX(nn.Module):
+
+    def __init__(self,
+                 in_channels: int,
+                 out_channels: int):
+        super(MBConvX, self).__init__()
+
+        # self._use_se = True
+
+        inner_channels = in_channels * 6
+
+        ex_attrs = dict(
+            in_channels=in_channels, out_channels=inner_channels,
+            kernel_size=1, stride=1, groups=1, bias=False,
+            bn_momentum=0.9, bn_eps=1e-5)
+        dw_attrs = dict(
+            in_channels=inner_channels, out_channels=inner_channels,
+            kernel_size=3, stride=1, groups=inner_channels, bias=False,
+            bn_momentum=0.9, bn_eps=1e-5)
+        op_attrs = dict(
+            in_channels=inner_channels, out_channels=out_channels,
+            kernel_size=1, stride=1, groups=1, bias=False,
+            bn_momentum=0.9, bn_eps=1e-5)
+
+        self.conv_ip = ConvBNR(**ex_attrs)
+        self.conv_dw = ConvBNR(**dw_attrs)
+        self.conv_op = ConvBNR(use_relu=False, **op_attrs)
+
+    def forward(self, inputs):
+        x = inputs
+
+        return x
+
+
+class EfficientNetBase(nn.Module):
+
+    def __init__(self, ):
+        pass
+
+
+if __name__ == '__main__':
+    e = EfficientNetBase()
+    print(vars(e))
+