\;jH#ddlZddlmcmZddlmZddlmZddlmZm Z m Z m Z m Z m Z ddlmZddlmZgZddiZd ZGd d ej*ZGd d ej*ZGddej*ZddZy)N)nn) ParamAttr)AdaptiveAvgPool2D AvgPool2DConv2DDropoutLinear MaxPool2D)Uniform)get_weights_path_from_url googlenet)zWhttps://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GoogLeNet_pretrained.pdparams 80c06f038e905c53ab32c40eca6e26aecPd|dz|zz dz}tt| |}|S)Ng@g?) initializer)rr )channels filter_sizestdv param_attrs gG:\00. PROJECTS\API\Inventory\templateJSON\kerjaOCR\Lib\site-packages\paddle/vision/models/googlenet.pyxavierr(s3 ;>H, -# 5Dwud';s JJv r)rr__name__ __module__ __qualname__r%r. __classcell__r*s@rrr.sGH  rrc$eZdZfdZdZxZS) Inceptionc  t |t||d|_t||d|_t||d|_t||d|_t||d|_tddd|_ t||d|_ y)Nr)rr r!) r$r%r_conv1_conv3r_conv3_conv5r_conv5r _pool_convprj) r'input_channelsoutput_channelsfilter1filter3Rfilter3filter5Rfilter5projr*s rr%zInception.__init__Ds ;  1= '15  1= '15 1QB !.$: rcT|j|}|j|}|j|}|j|}|j |}|j |}|j |}tj||||gd} tj| } | S)Nraxis) r:r;r<r=r>r?r@paddleconcatFrelu) r'r,conv1conv3rconv3conv5rconv5poolconvprjcats rr.zInception.forwardZs F#f% F#f% F#zz&!--%mmUE5':CffSk rr/r4s@rr6r6Cs ;,rr6c*eZdZdZdfd ZdZxZS) GoogLeNetaoGoogLeNet (Inception v1) model architecture from `"Going Deeper with Convolutions" `_. Args: num_classes (int, optional): Output dim of last fc layer. If num_classes <= 0, last fc layer will not be defined. Default: 1000. with_pool (bool, optional): Use pool before the last fc layer or not. Default: True. Returns: :ref:`api_paddle_nn_Layer`. An instance of GoogLeNet (Inception v1) model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import GoogLeNet >>> # Build model >>> model = GoogLeNet() >>> x = paddle.rand([1, 3, 224, 224]) >>> out, out1, out2 = model(x) >>> print(out.shape, out1.shape, out2.shape) [1, 1000] [1, 1000] [1, 1000] c t|||_||_t dddd|_t dd|_t ddd|_t ddd|_ tddddd d d d |_ td d d d dd dd|_ td d dddd dd|_ tdddddddd|_tddd d d ddd|_tdddddd dd|_tddd ddd d d |_tddd ddd d d |_tddddddd d |_|r4t)d|_t-dd|_t-dd|_|dkDrt3dd|_t7d |t9d d!|_t dd d|_t7d"d t9d#d!|_t3d$d|_ t7d |t9d d!|_!t dd d|_"t7d"d t9d#d!|_#t3d$d|_$t7d |t9d d!|_%yy)%Nr8@r)rr r` i0ipi ii@i@ir9rg?downscale_in_infer)pmodei) weight_attriigffffff?)&r$r% num_classes with_poolrr&r r?_conv_1_conv_2r6_ince3a_ince3b_ince4a_ince4b_ince4c_ince4d_ince4e_ince5a_ince5br_pool_5r_pool_o1_pool_o2r_dropr r_fc_out_conv_o1_fc_o1_drop_o1_out1_conv_o2_fc_o2_drop_o2_out2)r'rnror*s rr%zGoogLeNet.__init__s@ &"q"a+ 1Q7  R+  S!,  c2r3BC  c3S"b"E  c3CRD  c3S"b"E  c3S"b"E  c3S"b"E  c3S"c3G  c3S"c3G  c3S"c3G ,Q/DL%!A>DM%!A>DM ? 3-ABDJ!kvdADL &c32DM taIDK#c0DEDMkvdAODJ&c32DM taIDK#c0DEDMkvdAODJ# rc|j|}|j|}|j|}|j|}|j|}|j |}|j |}|j|}|j |}|j|}|j|}|j|}|j|}|j|}|j|}|j|}|||}}}|jr3|j|}|j|}|j!|}|j"dkDr|j%|}t'j(|ddg}|j+|}|j-|}t'j.|dd}|j1|}t3j4|}|j7|}|j9|}|j;|}t'j.|dd}|j=|}|j?|}|jA|}|||gS)Nrrr8rJr) start_axis stop_axis)!r&r?rprqrrrsrtrurvrwrxryrzror{r|r}rnr~rLsqueezerrflattenrrNrOrrrrrr) r'r,xince4aince4dince5boutout1out2s rr.zGoogLeNet.forwards JJv  JJqM LLO LLO JJqM LLO LLO JJqMa LL  LLOa LL  JJqM LLOa &&4T >>,,s#C==&D==&D   a **S/C..Aq62C,,s#C==&D>>$1CD;;t$D66$>$1CD;;t$D==&D::d#DT4  r)iT)r0r1r2__doc__r%r.r3r4s@rrYrYks6/Pb.!rrYc tdi|}d}|rX|tvs J|dtt|dt|d}tj|}|j ||S)aGoogLeNet (Inception v1) model architecture from `"Going Deeper with Convolutions" `_. 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:`GoogLeNet `. Returns: :ref:`api_paddle_nn_Layer`. An instance of GoogLeNet (Inception v1) model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import googlenet >>> # Build model >>> model = googlenet() >>> # Build model and load imagenet pretrained weight >>> # model = googlenet(pretrained=True) >>> x = paddle.rand([1, 3, 224, 224]) >>> out, out1, out2 = model(x) >>> print(out.shape, out1.shape, out2.shape) [1, 1000] [1, 1000] [1, 1000] r zJ model do not have a pretrained model now, you should set pretrained=Falserr)rY model_urlsr rLloadset_dict) pretrainedkwargsmodelarch weight_pathparams rr r s<   E D J  _V] ^ _ / t Q D!1!!4  K( u Lr)F)rLpaddle.nn.functionalr functionalrNpaddle.base.param_attrr paddle.nnrrrrr r paddle.nn.initializerr paddle.utils.downloadr __all__rrLayerrr6rYr rrrrs| ,*;   *%%P{!{!|*r