# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve. # # 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. from __future__ import absolute_import from __future__ import division from __future__ import print_function import paddle import paddle.nn as nn import paddle.nn.functional as F from paddle.nn import Conv2D, Dropout from paddle.nn import AdaptiveAvgPool2D, MaxPool2D from paddle.fluid.param_attr import ParamAttr from paddle.utils.download import get_weights_path_from_url __all__ = [] model_urls = { 'squeezenet1_0': ('https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_0_pretrained.pdparams', '30b95af60a2178f03cf9b66cd77e1db1'), 'squeezenet1_1': ('https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_1_pretrained.pdparams', 'a11250d3a1f91d7131fd095ebbf09eee'), } class MakeFireConv(nn.Layer): def __init__(self, input_channels, output_channels, filter_size, padding=0): super(MakeFireConv, self).__init__() self._conv = Conv2D(input_channels, output_channels, filter_size, padding=padding, weight_attr=ParamAttr(), bias_attr=ParamAttr()) def forward(self, x): x = self._conv(x) x = F.relu(x) return x class MakeFire(nn.Layer): def __init__(self, input_channels, squeeze_channels, expand1x1_channels, expand3x3_channels): super(MakeFire, self).__init__() self._conv = MakeFireConv(input_channels, squeeze_channels, 1) self._conv_path1 = MakeFireConv(squeeze_channels, expand1x1_channels, 1) self._conv_path2 = MakeFireConv(squeeze_channels, expand3x3_channels, 3, padding=1) def forward(self, inputs): x = self._conv(inputs) x1 = self._conv_path1(x) x2 = self._conv_path2(x) return paddle.concat([x1, x2], axis=1) class SqueezeNet(nn.Layer): """SqueezeNet model from `"SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size" `_. Args: version (str): Version of SqueezeNet, which can be "1.0" or "1.1". num_classes (int, optional): Output dim of last fc layer. If num_classes <= 0, last fc layer will not be defined. Default: 1000. with_pool (bool, optional): Use pool before the last fc layer or not. Default: True. Returns: :ref:`api_paddle_nn_Layer`. An instance of SqueezeNet model. Examples: .. code-block:: python import paddle from paddle.vision.models import SqueezeNet # build v1.0 model model = SqueezeNet(version='1.0') # build v1.1 model # model = SqueezeNet(version='1.1') x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] """ def __init__(self, version, num_classes=1000, with_pool=True): super(SqueezeNet, self).__init__() self.version = version self.num_classes = num_classes self.with_pool = with_pool supported_versions = ['1.0', '1.1'] assert version in supported_versions, \ "supported versions are {} but input version is {}".format( supported_versions, version) if self.version == "1.0": self._conv = Conv2D(3, 96, 7, stride=2, weight_attr=ParamAttr(), bias_attr=ParamAttr()) self._pool = MaxPool2D(kernel_size=3, stride=2, padding=0) self._conv1 = MakeFire(96, 16, 64, 64) self._conv2 = MakeFire(128, 16, 64, 64) self._conv3 = MakeFire(128, 32, 128, 128) self._conv4 = MakeFire(256, 32, 128, 128) self._conv5 = MakeFire(256, 48, 192, 192) self._conv6 = MakeFire(384, 48, 192, 192) self._conv7 = MakeFire(384, 64, 256, 256) self._conv8 = MakeFire(512, 64, 256, 256) else: self._conv = Conv2D(3, 64, 3, stride=2, padding=1, weight_attr=ParamAttr(), bias_attr=ParamAttr()) self._pool = MaxPool2D(kernel_size=3, stride=2, padding=0) self._conv1 = MakeFire(64, 16, 64, 64) self._conv2 = MakeFire(128, 16, 64, 64) self._conv3 = MakeFire(128, 32, 128, 128) self._conv4 = MakeFire(256, 32, 128, 128) self._conv5 = MakeFire(256, 48, 192, 192) self._conv6 = MakeFire(384, 48, 192, 192) self._conv7 = MakeFire(384, 64, 256, 256) self._conv8 = MakeFire(512, 64, 256, 256) self._drop = Dropout(p=0.5, mode="downscale_in_infer") self._conv9 = Conv2D(512, num_classes, 1, weight_attr=ParamAttr(), bias_attr=ParamAttr()) self._avg_pool = AdaptiveAvgPool2D(1) def forward(self, inputs): x = self._conv(inputs) x = F.relu(x) x = self._pool(x) if self.version == "1.0": x = self._conv1(x) x = self._conv2(x) x = self._conv3(x) x = self._pool(x) x = self._conv4(x) x = self._conv5(x) x = self._conv6(x) x = self._conv7(x) x = self._pool(x) x = self._conv8(x) else: x = self._conv1(x) x = self._conv2(x) x = self._pool(x) x = self._conv3(x) x = self._conv4(x) x = self._pool(x) x = self._conv5(x) x = self._conv6(x) x = self._conv7(x) x = self._conv8(x) if self.num_classes > 0: x = self._drop(x) x = self._conv9(x) if self.with_pool: x = F.relu(x) x = self._avg_pool(x) x = paddle.squeeze(x, axis=[2, 3]) return x def _squeezenet(arch, version, pretrained, **kwargs): model = SqueezeNet(version, **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.set_dict(param) return model def squeezenet1_0(pretrained=False, **kwargs): """SqueezeNet v1.0 model from `"SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size" `_. Args: pretrained (bool, optional): Whether to load pre-trained weights. If True, returns a model pre-trained on ImageNet. Default: False. **kwargs (optional): Additional keyword arguments. For details, please refer to :ref:`SqueezeNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of SqueezeNet v1.0 model. Examples: .. code-block:: python import paddle from paddle.vision.models import squeezenet1_0 # build model model = squeezenet1_0() # build model and load imagenet pretrained weight # model = squeezenet1_0(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] """ return _squeezenet('squeezenet1_0', '1.0', pretrained, **kwargs) def squeezenet1_1(pretrained=False, **kwargs): """SqueezeNet v1.1 model from `"SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size" `_. Args: pretrained (bool, optional): Whether to load pre-trained weights. If True, returns a model pre-trained on ImageNet. Default: False. **kwargs (optional): Additional keyword arguments. For details, please refer to :ref:`SqueezeNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of SqueezeNet v1.1 model. Examples: .. code-block:: python import paddle from paddle.vision.models import squeezenet1_1 # build model model = squeezenet1_1() # build model and load imagenet pretrained weight # model = squeezenet1_1(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] """ return _squeezenet('squeezenet1_1', '1.1', pretrained, **kwargs)