hello-world alexnet dygraph commit-20200612

dygraph/alexnet/network.py

+"""
+动态图构建 AlexNet
+"""
+import paddle.fluid as fluid
+import numpy as np
+
+class Conv2D(fluid.dygraph.Layer):
+    def __init__(self,
+                 name_scope,
+                 num_channels,
+                 num_filters,
+                 filter_size,
+                 stride=1,
+                 padding=0,
+                 dilation=1,
+                 groups=1,
+                 act=None,
+                 use_cudnn=False,
+                 param_attr=None,
+                 bias_attr=None):
+
+        super(Conv2D, self).__init__(name_scope)
+
+        self._conv2d = fluid.dygraph.Conv2D(
+            num_channels=num_channels,
+            num_filters=num_filters,
+            filter_size=filter_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            param_attr=param_attr,
+            bias_attr=bias_attr,
+            act=act,
+            use_cudnn=use_cudnn)
+
+    def forward(self, inputs):
+        x = self._conv2d(inputs)
+        return x
+
+class Conv2DPool(fluid.dygraph.Layer):
+    def __init__(self,
+                name_scope,
+                num_channels,
+                num_filters,
+                filter_size,
+                pool_size,
+                pool_stride,
+                pool_padding=0,
+                pool_type='max',
+                global_pooling=False,
+                conv_stride=1,
+                conv_padding=0,
+                conv_dilation=1,
+                conv_groups=1,
+                act=None,
+                use_cudnn=False,
+                param_attr=None,
+                bias_attr=None):
+
+        super(Conv2DPool, self).__init__(name_scope)
+
+        self._conv2d = fluid.dygraph.Conv2D(
+            num_channels=num_channels,
+            num_filters=num_filters,
+            filter_size=filter_size,
+            stride=conv_stride,
+            padding=conv_padding,
+            dilation=conv_dilation,
+            groups=conv_groups,
+            param_attr=param_attr,
+            bias_attr=bias_attr,
+            act=act,
+            use_cudnn=use_cudnn)
+
+        self._pool2d = fluid.dygraph.Pool2D(
+            pool_size=pool_size,
+            pool_type=pool_type,
+            pool_stride=pool_stride,
+            pool_padding=pool_padding,
+            global_pooling=global_pooling,
+            use_cudnn=use_cudnn)
+
+    def forward(self, inputs):
+        x = self._conv2d(inputs)
+        x = self._pool2d(x)
+        return x
+
+class AlexNet(fluid.dygraph.Layer):
+    def __init__(self, name_scope, class_dim):
+        super(AlexNet, self).__init__(name_scope)
+
+        self.conv_pool_1 = Conv2DPool(self.full_name(), 3, 64, 11, 3, 2, conv_stride=4, conv_padding=2, act='relu')
+        self.conv_pool_2 = Conv2DPool(self.full_name(), 64, 192, 5, 3, 2, conv_stride=1, conv_padding=2, act='relu')
+        self.conv_3 = Conv2D(self.full_name(), 192, 384, 3, 1, 1, act='relu')
+        self.conv_4 = Conv2D(self.full_name(), 384, 256, 3, 1, 1, act='relu')
+        self.conv_pool_5 = Conv2DPool(self.full_name(), 256, 256, 3, 3, 2, conv_stride=1, conv_padding=1, act='relu')
+        self.fc6 = fluid.dygraph.FC(self.full_name(), 9216, 4096, act='relu')
+        self.fc7 = fluid.dygraph.FC(self.full_name(), 4096, 4096, act='relu')
+        self.fc8 = fluid.dygraph.FC(self.full_name(), 4096, class_dim, act='softmax')
+
+    def forward(self, inputs, label=None):
+        out = self.conv_pool_1(inputs)
+        out = self.conv_pool_2(out)
+        out = self.conv_3(out)
+        out = self.conv_4(out)
+        out = self.conv_pool_5(out)
+        out = self.fc6(out)
+        out = fluid.layers.dropout(out, 0.5)
+        out = self.fc7(out)
+        out = fluid.layers.dropout(out, 0.5)
+        out = self.fc8(out)
+
+        if label is not None:
+            acc = fluid.layers.accuracy(input=out, label=label)
+            return out, acc
+        else:
+            return out
+
+if __name__ == '__main__':
+    with fluid.dygraph.guard():
+        alexnet = AlexNet('alex-net', 3)
+        img = np.zeros([2, 3, 224, 224]).astype('float32')
+        img = fluid.dygraph.to_variable(img)
+        outs = alexnet(img).numpy()
+
+        print(outs)
+
+