o QeB@sNddlmZddlmZddlmZddlZddlZddlmZddlmZm Z m Z m Z ddlm Z m Z mZddlmZddlmZdd lmZgZd d d d ddZGdddejZGdddejZGdddejZGdddejZGdddejZGdddejZddZd)dd Zd)d!d"Z d)d#d$Z!d)d%d&Z"d)d'd(Z#dS)*)absolute_import)division)print_functionN)Conv2D BatchNormLinearDropout)AdaptiveAvgPool2D MaxPool2D AvgPool2D)Uniform) ParamAttr)get_weights_path_from_url)zYhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet121_pretrained.pdparamsZ db1b239ed80a905290fd8b01d3af08e4)zYhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet161_pretrained.pdparamsZ 62158869cb315098bd25ddbfd308a853)zYhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet169_pretrained.pdparamsZ 82cc7c635c3f19098c748850efb2d796)zYhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet201_pretrained.pdparamsZ 16ca29565a7712329cf9e36e02caaf58)zYhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet264_pretrained.pdparamsZ 3270ce516b85370bba88cfdd9f60bff4) densenet121 densenet161 densenet169 densenet201 densenet264c.eZdZ    dfdd ZddZZS) BNACConvLayerrreluc s<tt|t||d|_t||||||tdd|_dS)NactFZ in_channelsZ out_channels kernel_sizestridepaddinggroups weight_attr bias_attr)superr__init__r _batch_normrr _convself num_channels num_filters filter_sizerpadrr __class__MD:\Projects\ConvertPro\env\Lib\site-packages\paddle/vision/models/densenet.pyr"4s zBNACConvLayer.__init__cC||}||}|SN)r#r$r&inputyr-r-r.forwardH  zBNACConvLayer.forwardrrrr__name__ __module__ __qualname__r"r4 __classcell__r-r-r+r.r2rc$eZdZfddZddZZS) DenseLayercs^tt|||_t|||dddd|_t|||dddd|_|r-t|dd|_dSdS)Nrrr'r(r)r*rZdownscale_in_infer)pmode) r!r>r"dropoutr bn_ac_func1 bn_ac_func2r dropout_func)r&r' growth_ratebn_sizerCr+r-r.r"Ps"zDenseLayer.__init__cCs:||}||}|jr||}tj||gdd}|S)Nr)Zaxis)rDrErCrFpaddleconcat)r&r2convr-r-r.r4cs   zDenseLayer.forwardr7r-r-r+r.r>Ns r>cs(eZdZ dfdd ZddZZS) DenseBlockNc sdtt|||_g|_|}t|D]}|j|d||dt ||||d||}qdS)Nz{}_{}r)r'rGrHrC) r!rLr"rCdense_layer_funcrangeappend add_sublayerformatr>) r&r' num_layersrHrGrCnameZ pre_channellayerr+r-r.r"ns"  zDenseBlock.__init__cCs|}|jD]}||}q|Sr0)rM)r&r2rKfuncr-r-r.r4s  zDenseBlock.forwardr0r7r-r-r+r.rLlsrLcr=)TransitionLayercs6tt|t||dddd|_tdddd|_dS)Nrrr?rrr)r!rVr"r conv_ac_funcr pool2d_avg)r&r'num_output_featuresr+r-r.r"szTransitionLayer.__init__cCr/r0)rYrZr1r-r-r.r4r5zTransitionLayer.forwardr7r-r-r+r.rVs  rVcr) ConvBNLayerrrrc s<tt|t||||||tdd|_t||d|_dS)NFrr)r!r\r"rr r$rr#r%r+r-r.r"szConvBNLayer.__init__cCr/r0)r$r#r1r-r-r.r4r5zConvBNLayer.forwardr6r7r-r-r+r.r\r<r\cs4eZdZdZ     d fdd Zd d ZZS) DenseNetaDenseNet model from `"Densely Connected Convolutional Networks" `_. Args: layers (int, optional): Layers of DenseNet. Default: 121. bn_size (int, optional): Expansion of growth rate in the middle layer. Default: 4. dropout (float, optional): Dropout rate. Default: :math:`0.0`. num_classes (int, optional): Output dim of last fc layer. If num_classes <= 0, last fc layer will not be defined. Default: 1000. with_pool (bool, optional): Use pool before the last fc layer or not. Default: True. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet model. Examples: .. code-block:: python import paddle from paddle.vision.models import DenseNet # build model densenet = DenseNet() x = paddle.rand([1, 3, 224, 224]) out = densenet(x) print(out.shape) # [1, 1000] yTcstt|||_||_gd}||vsJd||ddgdfddgdfddgd fddgd fddgd fd}||\}} } td |d dd dd|_td ddd|_ | |_ g|_ g|_ |} |} t | D]N\} }|j |d| dt| ||| |dt| dd| || } | } | t| dkr|j |d| dt| | dd| d} | d} qit| dd|_|jrtd|_|jdkrdt| d}t| |tt| |dtd|_dSdS)N)r^z-supported layers are {} but input layer is {}@ ) `0)rhri$rj)rhrirgrg)rhrirmrg)rhrirfrmr@rWr)r'r(r)rr*rrrXz db_conv_{}rK)r'rRrHrGrCrSz tr_conv{}_blk)r'r[rrg?)Z initializer)rr )r!r]r" num_classes with_poolrQr\ conv1_funcr pool2d_max block_configdense_block_func_listtransition_func_list enumeraterOrPrLstrlenrVr batch_normr rZmathsqrtrr r out)r&layersrHrCrprqZsupported_layersZ densenet_specZnum_init_featuresrGrtZpre_num_channelsZ num_featuresirRZstdvr+r-r.r"s            zDenseNet.__init__cCs||}||}t|jD]\}}|j||}|t|jdkr*|j||}q||}|jr8| |}|j dkrJt j |ddd}| |}|S)Nrr)Z start_axisZ stop_axis)rrrsrwrtruryrvrzrqrZrprIflattenr})r&r2rKrrRr3r-r-r.r4s      zDenseNet.forward)r^r_r`raT)r8r9r:__doc__r"r4r;r-r-r+r.r]sEr]cKs^tdd|i|}|r-|tvsJd|tt|dt|d}t|}|||S)Nr~zL{} model do not have a pretrained model now, you should set pretrained=Falserrr-)r] model_urlsrQrrIloadZset_dict)archr~ pretrainedkwargsmodelZ weight_pathparamr-r-r. _densenet3s    rFcKtdd|fi|S)aDenseNet 121-layer model from `"Densely Connected Convolutional Networks" `_. 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:`DenseNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet 121-layer model. Examples: .. code-block:: python import paddle from paddle.vision.models import densenet121 # build model model = densenet121() # build model and load imagenet pretrained weight # model = densenet121(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] rr^rrrr-r-r.rArcKr)a'DenseNet 161-layer model from `"Densely Connected Convolutional Networks" `_. 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:`DenseNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet 161-layer model. Examples: .. code-block:: python from paddle.vision.models import densenet161 # build model model = densenet161() # build model and load imagenet pretrained weight # model = densenet161(pretrained=True) rrbrrr-r-r.rbsrcKr)aDenseNet 169-layer model from `"Densely Connected Convolutional Networks" `_. 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:`DenseNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet 169-layer model. Examples: .. code-block:: python import paddle from paddle.vision.models import densenet169 # build model model = densenet169() # build model and load imagenet pretrained weight # model = densenet169(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] rrcrrr-r-r.r|rrcKr)aDenseNet 201-layer model from `"Densely Connected Convolutional Networks" `_. 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:`DenseNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet 201-layer model. Examples: .. code-block:: python import paddle from paddle.vision.models import densenet201 # build model model = densenet201() # build model and load imagenet pretrained weight # model = densenet201(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] rrdrrr-r-r.rsrcKr)aDenseNet 264-layer model from `"Densely Connected Convolutional Networks" `_. 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:`DenseNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of DenseNet 264-layer model. Examples: .. code-block:: python import paddle from paddle.vision.models import densenet264 # build model model = densenet264() # build model and load imagenet pretrained weight # model = densenet264(pretrained=True) x = paddle.rand([1, 3, 224, 224]) out = model(x) print(out.shape) # [1, 1000] rrerrr-r-r.rrr)F)$ __future__rrrr{rIZ paddle.nnnnrrrrr r r Zpaddle.nn.initializerr Zpaddle.fluid.param_attrr Zpaddle.utils.downloadr__all__rZLayerrr>rLrVr\r]rrrrrrr-r-r-r.s<       y  !  !