o #j/K@sddlZddlZddlmZddlmZddlmZmZmZm Z m Z ddl m Z ddl mZddlmZgZd d iZGd d d ejZGd ddejZGdddejZGdddejZGdddejZGdddejZGdddejZddd ZdS)N)nn) ParamAttr)AdaptiveAvgPool2D AvgPool2DDropoutLinear MaxPool2D)Uniform)get_weights_path_from_url)ConvNormActivation inception_v3)z>https://paddle-hapi.bj.bcebos.com/models/inception_v3.pdparamsZ 649a4547c3243e8b59c656f41fe330b8c$eZdZfddZddZZS) InceptionStemcsttdddddtjd|_tdddddtjd|_tddddtjd|_tdddd |_ tdd ddtjd|_ td d ddtjd|_ dS) N r r in_channels out_channels kernel_sizestridepaddingactivation_layer@rrrrr)rrrP) super__init__r rReLU conv_1a_3x3 conv_2a_3x3 conv_2b_3x3rmax_pool conv_3b_1x1 conv_4a_3x3)self __class__a/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddle/vision/models/inceptionv3.pyr%sN  zInceptionStem.__init__cCsJ||}||}||}||}||}||}||}|S)N)r!r"r#r$r%r&)r'xr*r*r+forwardOs       zInceptionStem.forward__name__ __module__ __qualname__rr- __classcell__r*r*r(r+r$s *rcr) InceptionAcstt|dddtjd|_t|dddtjd|_tddddtjd|_t|dddtjd|_tddd dtjd|_ tddd dtjd|_ t d ddd d |_ t||ddtjd|_ dS) Nrrrr0r `rFrrrZ exclusive)rrr rr branch1x1 branch5x5_1 branch5x5_2branch3x3dbl_1branch3x3dbl_2branch3x3dbl_3r branch_poolbranch_pool_conv)r' num_channelsZ pool_featuresr(r*r+r[sj  zInceptionA.__init__cCsj||}||}||}||}||}||}||}||}tj ||||gdd}|SNr)Zaxis) r8r9r:r;r<r=r>r?paddleconcat)r'r,r8Z branch5x5 branch3x3dblr>r*r*r+r-s         zInceptionA.forwardr.r*r*r(r+r3Zs :r3cr) InceptionBcsxtt|ddddtjd|_t|dddtjd|_tdd ddtjd|_td d dddtjd|_t ddd |_ dS) Nrr rrrrrr6rr) rrr rr branch3x3r;r<r=rr>r'r@r(r*r+rs@   zInceptionB.__init__cCsJ||}||}||}||}||}tj|||gdd}|SrA)rHr;r<r=r>rBrC)r'r,rHrDr>r*r*r+r-s     zInceptionB.forwardr.r*r*r(r+rEs $rEcr) InceptionCcstt|dddtjd|_t||dddtjd|_t||dddtjd|_t|dddd tjd|_t||ddtjd|_ t||dd tjd|_ t||ddtjd|_ t||dd tjd|_ t|dddtjd|_ td ddd d |_t|dddtjd|_dS) NrrrrrrrrrLrrrrFr7)rrr rr r8 branch7x7_1 branch7x7_2 branch7x7_3branch7x7dbl_1branch7x7dbl_2branch7x7dbl_3branch7x7dbl_4branch7x7dbl_5rr>r?)r'r@Z channels_7x7r(r*r+rs   zInceptionC.__init__cCs||}||}||}||}||}||}||}||}||}| |}| |}t j ||||gdd}|SrA) r8rPrQrRrSrTrUrVrWr>r?rBrC)r'r,r8Z branch7x7Z branch7x7dblr>r*r*r+r-,s            zInceptionC.forwardr.r*r*r(r+rJs RrJcr) InceptionDcstt|dddtjd|_tdddddtjd|_t|dddtjd|_tddd d tjd|_tddd d tjd|_ tdddddtjd|_ t ddd |_ dS)Nrrrr@rr rrKrMrNrOrG) rrr rr branch3x3_1 branch3x3_2 branch7x7x3_1 branch7x7x3_2 branch7x7x3_3 branch7x7x3_4rr>rIr(r*r+rDs\   zInceptionD.__init__cCs^||}||}||}||}||}||}||}tj|||gdd}|SrA) rZr[r\r]r^r_r>rBrC)r'r,rHZ branch7x7x3r>r*r*r+r-vs       zInceptionD.forwardr.r*r*r(r+rXCs 2rXcr) InceptionEcstt|dddtjd|_t|dddtjd|_tddddtjd|_tdddd tjd|_t|d ddtjd|_ td dd dtjd|_ tddddtjd|_ tdddd tjd|_ t d ddd d |_t|dddtjd|_dS)NrYrrrrF)rr)rr)rr)rrirFr7r)rrr rr r8rZ branch3x3_2a branch3x3_2br;r<branch3x3dbl_3abranch3x3dbl_3brr>r?rIr(r*r+rs  zInceptionE.__init__cCs||}||}||||g}tj|dd}||}||}||| |g}tj|dd}| |}| |}tj||||gdd}|SrA) r8rZrarbrBrCr;r<rcrdr>r?)r'r,r8rHrDr>r*r*r+r-s$       zInceptionE.forwardr.r*r*r(r+r`s Gr`cs*eZdZdZdfdd ZddZZS) InceptionV3a"Inception v3 model from `"Rethinking the Inception Architecture for Computer Vision" `_. 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 Inception v3 model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import InceptionV3 >>> inception_v3 = InceptionV3() >>> x = paddle.rand([1, 3, 299, 299]) >>> out = inception_v3(x) >>> print(out.shape) [1, 1000] Tcst||_||_gdgdgdggdgdgdgddgd |_|jd }|jd }|jd }|jd }|jd}t|_t|_ t t |dD]}t |d||d|} |j | qNt t |D]}t||} |j | qjt t |dD]}t|d||d|} |j | qt t |D]}t||} |j | qt t |D]}t||} |j | q|rtd|_|dkrtddd|_dtd}td|tt| |dtd|_dSdS)N)r )rrrrh)ririri)rkrrii) inception_a inception_b inception_c inception_d inception_ermrornrprqrrg?Zdownscale_in_infer)pmodeg?g@)Z initializer)Z weight_attrZ bias_attr)rr num_classes with_poolZ layers_configrinception_stemrZ LayerListinception_block_listrangelenr3appendrErJrXr`ravg_poolrdropoutmathsqrtrrr fc)r'rtruZinception_a_listZinception_c_listZinception_b_listZinception_d_listZinception_e_listirmrnrorprqZstdvr(r*r+rs^            zInceptionV3.__init__cCsb||}|jD]}||}q|jr||}|jdkr/tj|ddgd}||}||}|S)Nrrl)shape) rvrwrur{rtrBZreshaper|r)r'r,Zinception_blockr*r*r+r-=s       zInceptionV3.forward)rfT)r/r0r1__doc__rr-r2r*r*r(r+res:reFcKs^tdi|}d}|r-|tvsJ|dtt|dt|d}t|}|||S)aInception v3 model from `"Rethinking the Inception Architecture for Computer Vision" `_. 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:`InceptionV3 `. Returns: :ref:`api_paddle_nn_Layer`. An instance of Inception v3 model. Examples: .. code-block:: python >>> import paddle >>> from paddle.vision.models import inception_v3 >>> # Build model >>> model = inception_v3() >>> # Build model and load imagenet pretrained weight >>> # model = inception_v3(pretrained=True) >>> x = paddle.rand([1, 3, 299, 299]) >>> out = model(x) >>> print(out.shape) [1, 1000] r zJ model do not have a pretrained model now, you should set pretrained=FalserrNr*)re model_urlsr rBloadZset_dict)Z pretrainedkwargsmodelarchZ weight_pathparamr*r*r+r Ls   )F)r}rBrZpaddle.base.param_attrrZ paddle.nnrrrrrZpaddle.nn.initializerr Zpaddle.utils.downloadr opsr __all__rZLayerrr3rErJrXr`rer r*r*r*r+s&     6L3jBcd