\;jgddlZddlmZddlmZddlmZddlmZgZdd iZ Gd d ejZ Gd d ejZ ddZ ddZy)N)nn)get_weights_path_from_url)ConvNormActivation)_make_divisiblezmobilenetv2_1.0)zChttps://paddle-hapi.bj.bcebos.com/models/mobilenet_v2_x1.0.pdparams 0340af0a901346c8d46f4529882fb63dc<eZdZejffd ZdZxZS)InvertedResidualc t|||_|dvsJtt ||z}|jdk(xr||k(|_g}|dk7r-|j t||d|tj|jt|||||tjtj||dddd||gtj||_ y)N)rrr kernel_size norm_layeractivation_layer)stridegroupsrrrF) bias_attr)super__init__rintrounduse_res_connectappendrrReLU6extendConv2D Sequentialconv) selfinpoupr expand_ratior hidden_dimlayers __class__s iG:\00. PROJECTS\API\Inventory\templateJSON\kerjaOCR\Lib\site-packages\paddle/vision/models/mobilenetv2.pyrzInvertedResidual.__init__!s  s\123 #{{a/>C3J 1  MM" !)%'XX    "!%)%'XX  *c1aeD3  MM6* cd|jr||j|zS|j|S)N)rrrxs r&forwardzInvertedResidual.forwardFs,   tyy|# #99Q< r')__name__ __module__ __qualname__r BatchNorm2Drr+ __classcell__r%s@r&r r s9;#+J r'r c*eZdZdZdfd ZdZxZS) MobileNetV2a[MobileNetV2 model from `"MobileNetV2: Inverted Residuals and Linear Bottlenecks" `_. Args: scale (float, optional): Scale of channels in each layer. Default: 1.0. 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 MobileNetV2 model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import MobileNetV2 >>> model = MobileNetV2() >>> x = paddle.rand([1, 3, 224, 224]) >>> out = model(x) >>> print(out.shape) [1, 1000] c lt|||_||_d}d}t}d}t j }gdgdgdgdgdgd gd g} t||z|}t|td |z||_ td |d |t jg} | D]M\} } } }t| |z|}t| D])}|dk(r|nd}| j||||| ||}+O| jt||jd|t jt j| |_|rt j d|_|jdkDrMt jt j$dt j&|j||_yy)N i)rrr)rr)r8r5r)r8@r)r8`r:r)r8r:r)r8i@rr?r:r)rrrrr)r"rr g?)rr num_classes with_poolr rr/rmax last_channelrrrangerrfeaturesAdaptiveAvgPool2D pool2d_avgDropoutLinear classifier)rscaler@rA input_channelrCblock round_nearestrinverted_residual_settingrEtcnsoutput_channelirr%s r&rzMobileNetV2.__init__is &"    ^^       % !( (=}M + 3sE? *M  %!#   4JAq!Q,QY FN1X1f!%&%&#- !/ 4  !!%!#    x0  2215DO   a  mm 34+<+>"A   a q!$A"Ar')r?iT)r,r-r.__doc__rr+r0r1s@r&r3r3Ms6CJ r'r3c tdi|}|rX|tvs J|dtt|dt|d}tj|}|j ||S)NzJ model do not have a pretrained model now, you should set pretrained=Falserr)r3 model_urlsrrWload load_dict)arch pretrainedkwargsmodel weight_pathparams r& _mobilenetresx  !& !E J  _V] ^ _ / t Q D!1!!4  K(  Lr'c <tdt|z|fd|i|}|S)aMobileNetV2 from `"MobileNetV2: Inverted Residuals and Linear Bottlenecks" `_. Args: pretrained (bool, optional): Whether to load pre-trained weights. If True, returns a model pre-trained on ImageNet. Default: False. scale (float, optional): Scale of channels in each layer. Default: 1.0. **kwargs (optional): Additional keyword arguments. For details, please refer to :ref:`MobileNetV2 `. Returns: :ref:`api_paddle_nn_Layer`. An instance of MobileNetV2 model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import mobilenet_v2 >>> # Build model >>> model = mobilenet_v2() >>> # Build model and load imagenet pretrained weight >>> # model = mobilenet_v2(pretrained=True) >>> # Build mobilenet v2 with scale=0.5 >>> model = mobilenet_v2(scale=0.5) >>> x = paddle.rand([1, 3, 224, 224]) >>> out = model(x) >>> print(out.shape) [1, 1000] mobilenetv2_rK)restr)r`rKrarbs r& mobilenet_v2ris4B U#Z 7< @F E Lr')F)Fr?)rWrpaddle.utils.downloadropsr_utilsr__all__r\Layerr r3rerir[r'r&rosZ;$#  * rxx* Zj"((jZ  $r'