j# LddlmZddlmZmZddlZddlmZddlmcm Z ddl m Z ddl m Z ddlmZmZmZdd lmZdd lmZmZgd ZGd d ejZGddejZdedeedededef dZedddZGddeZGddeZ eedej!fddd deedededefd!Z"eede j!fddd dee dededefd"Z#dS)#)partial)AnyOptionalN)ImageClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) SqueezeNetSqueezeNet1_0_WeightsSqueezeNet1_1_Weights squeezenet1_0 squeezenet1_1c XeZdZdededededdf fd Zdejdejfd ZxZS) Fireinplanessqueeze_planesexpand1x1_planesexpand3x3_planesreturnNct||_tj||d|_tjd|_tj||d|_tjd|_ tj||dd|_ tjd|_ dS)Nr  kernel_sizeTinplace)rpadding) super__init__rnnConv2dsqueezeReLUsqueeze_activation expand1x1expand1x1_activation expand3x3expand3x3_activation)selfrrrr __class__s NC:\Projects\Historical\venv\Lib\site-packages\torchvision/models/squeezenet.pyr$z Fire.__init__s   y>qIII "$'$"7"7"7>3CQRSSS$&GD$9$9$9!>3CQR\]^^^$&GD$9$9$9!!!xc|||}tj||||||gdSNr )r)r'torchcatr+r*r-r,r.r2s r0forwardz Fire.forwardsv  # #DLLOO 4 4y  & &t~~a'8'8 9 94;T;TUYUcUcdeUfUf;g;g hjk   r1) __name__ __module__ __qualname__intr$r5Tensorr8 __classcell__r/s@r0rrs::c:S:dg:lp::::::  %,        r1rc VeZdZd dedededdffd Zd ejdejfd Z xZ S) r1_0?version num_classesdropoutrNctt|||_|dkrt jt jddddt jdt jddd tdd d d td d d d td d d d t jddd tdd d d tddddtddddtdd ddt jddd tdd dd |_ n(|dkrt jt jdd ddt jdt jddd td d d d td d d d t jddd td d d d tdd d d t jddd tddddtddddtdd ddtdd dd |_ ntd|dt jd|jd}t jt j ||t jdt j d|_|D]w}t!|tjr[||urt#j|jddnt#j|j|jt#j|jdxdS)NrAr!`r)rstrideTr)rrJ ceil_mode@ 0ii1_1zUnsupported SqueezeNet version z: 1_0 or 1_1 expectedr r)p)r r gg{Gz?)meanstdr)r#r$rrEr% Sequentialr&r( MaxPool2drfeatures ValueErrorDropoutAdaptiveAvgPool2d classifiermodules isinstanceinitnormal_weightkaiming_uniform_bias constant_)r.rDrErF final_convmr/s r0r$zSqueezeNet.__init__%sJ D!!!& e  M !RQq999%%% 1EEERR$$S"b"%%S"c3'' 1EEES"c3''S"c3''S"c3''S"c3'' 1EEES"c3''DMM  M !RQq999%%% 1EEERR$$S"b"%% 1EEES"c3''S"c3'' 1EEES"c3''S"c3''S"c3''S"c3''DMM&]w]]]^^ ^YsD$4!DDD - J ! ! !:rwt/D/D/DbFZ[aFbFb   . .A!RY'' . ??L>>>>>)!(3336%N161--- . .r1r2c||}||}tj|dSr4)rYr]r5flattenr7s r0r8zSqueezeNet.forward^s6 MM!   OOA  }Q"""r1)rArBrC) r9r:r;strr<floatr$r5r=r8r>r?s@r0rr$s7.7.7.#7.u7._c7.7.7.7.7.7.r##%,########r1rrDweightsprogresskwargsrc |)t|dt|jdt|fi|}|*|||d|S)NrE categoriesT)rm check_hash)rlenmetarload_state_dictget_state_dict)rDrlrmrnmodels r0 _squeezenetrwdss fmSl9S5T5TUUU w ) )& ) )E g44hSW4XXYYY Lr1z@https://github.com/pytorch/vision/pull/49#issuecomment-277560717zXThese weights reproduce closely the results of the paper using a simple training recipe.)rprecipe_docsc beZdZedeediedddddd id d d  ZeZdS)rz>https://download.pytorch.org/models/squeezenet1_0-b66bff10.pth crop_size)r~i  ImageNet-1Kg M@g{GT@zacc@1zacc@5gh|?5?g&1@min_size num_params_metrics_ops _file_sizeurl transformsrsN r9r:r;r rr _COMMON_META IMAGENET1K_V1DEFAULTr1r0rr|G L7.#>>>     !##       M"GGGr1rc beZdZedeediedddddd id d d  ZeZdS)rz>https://download.pytorch.org/models/squeezenet1_1-b8a52dc0.pthr{r|)ri(rgX9M@g-'T@rgtV?g"~@rrNrrr1r0rrrr1r pretrained)rlT)rlrmc Tt|}td||fi|S)aSqueezeNet model architecture from the `SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size `_ paper. Args: weights (:class:`~torchvision.models.SqueezeNet1_0_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.SqueezeNet1_0_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.squeezenet.SqueezeNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.SqueezeNet1_0_Weights :members: rA)rverifyrwrlrmrns r0rrs12$**733G ugx : :6 : ::r1c Tt|}td||fi|S)a/SqueezeNet 1.1 model from the `official SqueezeNet repo `_. SqueezeNet 1.1 has 2.4x less computation and slightly fewer parameters than SqueezeNet 1.0, without sacrificing accuracy. Args: weights (:class:`~torchvision.models.SqueezeNet1_1_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.SqueezeNet1_1_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.squeezenet.SqueezeNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.SqueezeNet1_1_Weights :members: rS)rrrwrs r0rrs16$**733G ugx : :6 : ::r1)$ functoolsrtypingrrr5torch.nnr% torch.nn.initr`transforms._presetsrutilsr_apir r r _metar _utilsrr__all__Modulerrrjboolrwrrrrrrrr1r0rs 555555''''''6666666666''''''BBBBBBBB m l l     29   $=#=#=#=#=#=#=#=#@  k "   $'P k K*K*,0E0S!TUUU26;;;./;BF;Y\;;;;VU;6,0E0S!TUUU26;;;./;BF;Y\;;;;VU;;;r1