Add DenseNet

python/paddle/tests/test_vision_models.py

@@ -82,6 +82,18 @@ def test_lenet(self):
         x = np.array(np.random.random((2, 1, 28, 28)), dtype=np.float32)
         lenet.predict_batch(x)
 
+    def test_densenet121(self):
+        self.models_infer('densenet121')
+
+    def test_densenet161(self):
+        self.models_infer('densenet161')
+
+    def test_densenet169(self):
+        self.models_infer('densenet169')
+
+    def test_densenet201(self):
+        self.models_infer('densenet201')
+
 
 if __name__ == '__main__':
     unittest.main()

python/paddle/vision/__init__.py

@@ -44,6 +44,10 @@
 from .models import vgg16  # noqa: F401
 from .models import vgg19  # noqa: F401
 from .models import LeNet  # noqa: F401
+from .models import densenet121  # noqa: F401
+from .models import densenet161  # noqa: F401
+from .models import densenet169  # noqa: F401
+from .models import densenet201  # noqa: F401
 from .transforms import BaseTransform  # noqa: F401
 from .transforms import Compose  # noqa: F401
 from .transforms import Resize  # noqa: F401

python/paddle/vision/models/__init__.py

@@ -28,6 +28,10 @@
 from .vgg import vgg16  # noqa: F401
 from .vgg import vgg19  # noqa: F401
 from .lenet import LeNet  # noqa: F401
+from .densenet import densenet121  # noqa: F401
+from .densenet import densenet161  # noqa: F401
+from .densenet import densenet169  # noqa: F401
+from .densenet import densenet201  # noqa: F401
 
 __all__ = [ #noqa
     'ResNet',
@@ -45,5 +49,9 @@
     'mobilenet_v1',
     'MobileNetV2',
     'mobilenet_v2',
-    'LeNet'
+    'LeNet',
+    'densenet121',
+    'densenet161',
+    'densenet169',
+    'densenet201'
 ]

python/paddle/vision/models/densenet.py

@@ -0,0 +1,156 @@
+# Copyright (c) 2020 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.
+
+import paddle
+import paddle.nn as nn
+
+from paddle.utils.download import get_weights_path_from_url
+
+__all__ = ['densenet121', 'densenet161', 'densenet169', 'densenet201']
+
+model_urls = {
+    'DenseNet121': (''),
+    'DenseNet161': (''),
+    'DenseNet169': (''),
+    'DenseNet201': ('')
+}
+
+
+class DenseLayer(nn.Layer):
+
+    def __init__(self, in_c, growth_rate, bn_size):
+        super().__init__()
+        out_c = growth_rate * bn_size
+        self.layers = nn.Sequential(
+            nn.BatchNorm2D(in_c),
+            nn.ReLU(),
+            nn.Conv2D(in_c, out_c, 1),
+            nn.BatchNorm2D(out_c),
+            nn.ReLU(),
+            nn.Conv2D(
+                out_c, growth_rate, 3, padding=1))
+
+    def forward(self, x):
+        y = self.layers(x)
+        return y
+
+class DenseBlock(nn.Layer):
+
+    def __init__(self, num_layers, in_c, growth_rate, bn_size):
+        super().__init__()
+        self.layers = nn.LayerList()
+        for ind in range(num_layers):
+            self.layers.append(
+                DenseLayer(
+                    in_c=in_c + ind * growth_rate,
+                    growth_rate=growth_rate,
+                    bn_size=bn_size))
+
+    def forward(self, x):
+        features = [x]
+        for layer in self.layers:
+            new_x = layer(paddle.concat(features, axis=1))
+            features.append(new_x)
+        return paddle.concat(features, axis=1)
+
+class Transition(nn.Layer):
+
+    def __init__(self, in_c, out_c):
+        super().__init__()
+        self.layers = nn.Sequential(
+            nn.BatchNorm2D(in_c),
+            nn.ReLU(), nn.Conv2D(in_c, out_c, 1), nn.AvgPool2D(2, 2))
+
+    def forward(self, x):
+        return self.layers(x)
+
+
+class DenseNet(nn.Layer):
+
+    def __init__(self,
+                 num_classes=1000,
+                 growth_rate=32,
+                 block=(6, 12, 24, 16),
+                 bn_size=4,
+                 out_c=64):
+        super().__init__()
+        self.conv_pool = nn.Sequential(
+                            nn.Conv2D(
+                                3, out_c, 7, stride=2, padding=3), nn.MaxPool2D(3, 2))
+        self.blocks = nn.LayerList()
+        in_c = out_c
+        for ind, n in enumerate(block):
+            self.blocks.append(DenseBlock(n, in_c, growth_rate, bn_size))
+            in_c += growth_rate * n
+            if ind != len(block) - 1:
+                self.blocks.append(Transition(in_c, in_c // 2))
+                in_c //= 2
+        self.blocks.append(
+            nn.Sequential(
+                nn.BatchNorm2D(in_c),
+                nn.ReLU(), nn.AdaptiveAvgPool2D((1, 1)), nn.Flatten()))
+        self.cls = nn.Linear(in_c, num_classes)
+
+    def forward(self, x):
+        x = self.conv_pool(x)
+        for layer in self.blocks:
+            x = layer(x)
+        x = self.cls(x)
+        return x
+
+def _DenseNet(arch, block_cfg, batch_norm, pretrained, **kwargs):
+    model = DenseNet(block=block_cfg, **kwargs)
+
+    if pretrained:
+        assert arch in model_urls, "{} model do not have a pretrained model now, you should set pretrained=False".format(
+            arch)
+        weight_path = get_weights_path_from_url(model_urls[arch][0],
+                                                model_urls[arch][1])
+
+        param = paddle.load(weight_path)
+        model.load_dict(param)
+
+    return model
+
+
+def densenet121(pretrained=False, batch_norm=False, **kwargs):
+    model_name = 'DenseNet121'
+    if batch_norm:
+        model_name += ('_bn')
+    return _DenseNet(model_name, (6, 12, 24, 16), batch_norm, pretrained, 
+                     **kwargs)
+
+
+def densenet161(pretrained=False, batch_norm=False, **kwargs):
+    model_name = 'DenseNet161'
+    if batch_norm:
+        model_name += ('_bn')
+    return _DenseNet(model_name, (6, 12, 32, 32), batch_norm, pretrained, 
+                     **kwargs)
+
+
+def densenet169(pretrained=False, batch_norm=False, **kwargs):
+    model_name = 'DenseNet169'
+    if batch_norm:
+        model_name += ('_bn')
+    return _DenseNet(model_name, (6, 12, 48, 32), batch_norm, pretrained, 
+                     **kwargs)
+
+
+def densenet201(pretrained=False, batch_norm=False, **kwargs):
+    model_name = 'DenseNet201'
+    if batch_norm:
+        model_name += ('_bn')
+    return _DenseNet(model_name, (6, 12, 64, 48), batch_norm, pretrained, 
+                     **kwargs)