o V hR@s2UddlZddlZddlmZddlmZddlmZmZm Z m Z m Z m Z m Z mZddlZddlmZmZddlmZddlmZmZdd lmZmZdd lmZd d lmZmZmZd d l m!Z!d dl"m#Z#m$Z$m%Z%gdZ&eGdddZ'Gddde'Z(Gddde'Z)Gdddej*Z+Gdddej*Z,Gdddej*Z-de ee(e)fde.de e/de ed e0d!ed"e-fd#d$Z1d%e2d!ed"e e ee(e)fe e/ffd&d'Z3d(e!iZ4e e2efe5d)<ie4d*d+d,Z6ie4d-d.d,Z7Gd/d0d0eZ8Gd1d2d2eZ9Gd3d4d4eZ:Gd5d6d6eZ;Gd7d8d8eZGd=d>d>eZ?Gd?d@d@eZ@GdAdBdBeZAGdCdDdDeZBee%dEe8jCfdFddGdHde e8d e0d!ed"e-fdIdJZDee%dEe9jCfdFddGdHde e9d e0d!ed"e-fdKdLZEee%dEe:jCfdFddGdHde e:d e0d!ed"e-fdMdNZFee%dEe;jCfdFddGdHde e;d e0d!ed"e-fdOdPZGee%dEejCfdFddGdHde e>d e0d!ed"e-fdUdVZJee%dEe?jCfdFddGdHde e?d e0d!ed"e-fdWdXZKee%dEe@jCfdFddGdHde e@d e0d!ed"e-fdYdZZLee%dEeAjCfdFddGdHde eAd e0d!ed"e-fd[d\ZMee%dEeBjCfdFddGdHde eBd e0d!ed"e-fd]d^ZNdS)_N) dataclass)partial)AnyCallableDictListOptionalSequenceTupleUnion)nnTensor)StochasticDepth)Conv2dNormActivationSqueezeExcitation)ImageClassificationInterpolationMode)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_make_divisible_ovewrite_named_paramhandle_legacy_interface) EfficientNetEfficientNet_B0_WeightsEfficientNet_B1_WeightsEfficientNet_B2_WeightsEfficientNet_B3_WeightsEfficientNet_B4_WeightsEfficientNet_B5_WeightsEfficientNet_B6_WeightsEfficientNet_B7_WeightsEfficientNet_V2_S_WeightsEfficientNet_V2_M_WeightsEfficientNet_V2_L_Weightsefficientnet_b0efficientnet_b1efficientnet_b2efficientnet_b3efficientnet_b4efficientnet_b5efficientnet_b6efficientnet_b7efficientnet_v2_sefficientnet_v2_mefficientnet_v2_lc @steZdZUeed<eed<eed<eed<eed<eed<edejfed<e dd ed ed e ed efddZ d S) _MBConvConfig expand_ratiokernelstrideinput_channels out_channels num_layers.blockNchannels width_mult min_valuereturncCst||d|S)N)r)r<r=r>rAS/var/www/vscode/kcb/lib/python3.10/site-packages/torchvision/models/efficientnet.pyadjust_channels8sz_MBConvConfig.adjust_channelsN) __name__ __module__ __qualname__float__annotations__intrr Module staticmethodrrCrArArArBr4.s &r4csreZdZ   ddedededededed ed ed eed ejfd dffdd Z e ded efddZ Z S) MBConvConfig?Nr5r6r7r8r9r:r= depth_multr;.r?c sL|||}|||}|||}| durt} t||||||| dSrD)rC adjust_depthMBConvsuper__init__) selfr5r6r7r8r9r:r=rOr; __class__rArBrS?s  zMBConvConfig.__init__cCstt||SrD)rJmathceil)r:rOrArArBrPRszMBConvConfig.adjust_depth)rNrNN) rErFrGrHrJrrr rKrSrLrP __classcell__rArArUrBrM=s6    rMcsPeZdZ d dededededededeed ejfd dffd d Z Z S)FusedMBConvConfigNr5r6r7r8r9r:r;.r?c s(|durt}t|||||||dSrD) FusedMBConvrRrS)rTr5r6r7r8r9r:r;rUrArBrSYs zFusedMBConvConfig.__init__rD) rErFrGrHrJrrr rKrSrYrArArUrBrZWs&  rZc s\eZdZefdedededejfdedejfddf fdd Z d e de fd d Z Z S) rQcnfstochastic_depth_prob norm_layer.se_layerr?Nc std|jkrdkstdtd|jdko"|j|jk|_g}tj}| |j|j }||jkrC| t |j|d||d| t |||j |j|||dtd|jd}| |||ttjddd | t ||jd|ddtj||_t|d |_|j|_dS) Nrrillegal stride value kernel_sizer^activation_layer)rbr7groupsr^rcT)inplace) activationrow)rRrSr7 ValueErrorr8r9use_res_connectr SiLUrCr5appendrr6maxr Sequentialr;rstochastic_depth) rTr\r]r^r_layersrcexpanded_channelssqueeze_channelsrUrArBrSisP        zMBConv.__init__inputcC&||}|jr||}||7}|SrDr;rjrorTrsresultrArArBforward  zMBConv.forward) rErFrGrrMrHrr rKrSr rxrYrArArUrBrQhs  :rQcsJeZdZdedededejfddffdd Zd e de fd d Z Z S) r[r\r]r^.r?Nc std|jkrdkstdtd|jdko"|j|jk|_g}tj}| |j|j }||jkrT| t |j||j |j||d| t ||jd|ddn| t |j|j|j |j||dtj||_t|d|_|j|_dS)Nrrr`rbr7r^rcrarh)rRrSr7rir8r9rjr rkrCr5rlrr6rnr;rro)rTr\r]r^rprcrqrUrArBrSsL      zFusedMBConv.__init__rscCrtrDrurvrArArBrxryzFusedMBConv.forward) rErFrGrZrHrr rKrSr rxrYrArArUrBr[s 4r[cseZdZ    ddeeeefdededede e d e j fd e ed dffd d Z ded efddZded efddZZS)r皙?Ninverted_residual_settingdropoutr] num_classesr^. last_channelr?c s0tt||stdt|trtdd|Ds!td|dur(tj }g}|dj }| t d|dd|tj d td d |D} d} |D]8} g} t| jD]&} t| }| rc|j|_ d |_|t| | }| ||||| d 7} qS| tj| qJ|d j}|dur|nd|}| t ||d |tj dtj||_td |_ttj|ddt|||_|D]V}t|tjrtjj |j!dd|j"durtj#|j"qt|tj tj$frtj%|j!tj#|j"qt|tjrdt&'|j(}tj)|j!| |tj#|j"qdS)a EfficientNet V1 and V2 main class Args: inverted_residual_setting (Sequence[Union[MBConvConfig, FusedMBConvConfig]]): Network structure dropout (float): The droupout probability stochastic_depth_prob (float): The stochastic depth probability num_classes (int): Number of classes norm_layer (Optional[Callable[..., nn.Module]]): Module specifying the normalization layer to use last_channel (int): The number of channels on the penultimate layer z1The inverted_residual_setting should not be emptycSsg|]}t|tqSrA) isinstancer4).0srArArB sz)EfficientNet.__init__..z:The inverted_residual_setting should be List[MBConvConfig]Nrrrzcss|]}|jVqdSrD)r:)rr\rArArB sz(EfficientNet.__init__..rreraT)prffan_out)moderN)*rRrSrrirr all TypeErrorr BatchNorm2dr8rlrrksumranger:copyr9r7rHr;rnfeaturesAdaptiveAvgPool2davgpoolDropoutLinear classifiermodulesConv2dinitkaiming_normal_weightbiaszeros_ GroupNormones_rWsqrt out_featuresuniform_)rTr}r~r]rr^rrpfirstconv_output_channelstotal_stage_blocksstage_block_idr\stage_ block_cnfsd_problastconv_input_channelslastconv_output_channelsm init_rangerUrArBrSs~           zEfficientNet.__init__xcCs.||}||}t|d}||}|S)Nr)rrtorchflattenrrTrrArArB _forward_implLs    zEfficientNet._forward_implcCs ||SrD)rrrArArBrxVs zEfficientNet.forward)r{r|NN)rErFrGr r rMrZrHrJrrr rKrSr rrxrYrArArUrBrs,c rr}r~rweightsprogresskwargsr?cKsT|durt|dt|jdt||fd|i|}|dur(||j|dd|S)Nr categoriesrT)r check_hash)rlenmetarload_state_dictget_state_dict)r}r~rrrrmodelrArArB _efficientnetZs rarchc KsB|drRtt|d|dd}|dddddd|d dd dd d |d d d d d d |d dd d dd|d d dddd|d d d ddd|d dddddg}d}||fS|drtdddd d d tddd d ddtddd dddtddd ddd td dddddtd dd dddg}d}||fS|drtdddd d dtddd d dd tddd ddd tddd dddtd dddddtd dd dd d!td ddd d"d g}d}||fS|d#rtddddddtddd dddtddd dd$dtddd d$dd%td dddd&d'td dd d&d(d)td ddd(d*dg}d}||fStd+|),Nefficientnet_br=rOr=rOrr r(Ppre@r10@ ir2i0ir3` izUnsupported model type ) startswithrrMpoprZri)rr bneck_confr}rrArArB_efficientnet_confmsZ  $     rr _COMMON_META)rrzUhttps://github.com/pytorch/vision/tree/main/references/classification#efficientnet-v1)min_sizerecipe)!rzUhttps://github.com/pytorch/vision/tree/main/references/classification#efficientnet-v2c @JeZdZedeeddejdieddddd id d d d dZ e Z dS)rzJhttps://download.pytorch.org/models/efficientnet_b0_rwightman-7f5810bc.pthrr crop_size resize_size interpolationidP ImageNet-1Kg?5^IlS@g5^I bW@zacc@1zacc@5gNbX9?g~jts4@1These weights are ported from the original paper. num_params_metrics_ops _file_size_docsurl transformsrN rErFrGrrrrBICUBIC_COMMON_META_V1 IMAGENET1K_V1DEFAULTrArArArBr* rc @seZdZedeeddejdieddddd id d d d dZ edeeddej diedddddd id ddddZ e Z dS)rzJhttps://download.pytorch.org/models/efficientnet_b1_rwightman-bac287d4.pthrrivrg +S@gClW@rgCl?gM">@rrrz@https://download.pytorch.org/models/efficientnet_b1-c27df63c.pthzOhttps://github.com/pytorch/vision/issues/3995#new-recipe-with-lr-wd-crop-tuninggʡS@gƻW@gA`">@$ These weights improve upon the results of the original paper by using a modified version of TorchVision's `new training recipe `_. )rrrrrrN) rErFrGrrrrrrrBILINEAR IMAGENET1K_V2rrArArArBrsR  rc @JeZdZedeeddejdiedddddid d d d d Z e Z dS)r zJhttps://download.pytorch.org/models/efficientnet_b2_rwightman-c35c1473.pthi rirgx&T@gp= W@rg rh?gʡEA@rrrNrrArArArBr rr c @r)r!zJhttps://download.pytorch.org/models/efficientnet_b3_rwightman-b3899882.pthi,rrirgnT@g~jtX@rgZd;?gd;OG@rrrNrrArArArBr!rr!c @r)r"zJhttps://download.pytorch.org/models/efficientnet_b4_rwightman-23ab8bcd.pthi|rri0!'rgjtT@gt&X@rg~jt@gKR@rrrNrrArArArBr"/rr"c @r)r#zJhttps://download.pytorch.org/models/efficientnet_b5_lukemelas-1a07897c.pthirirg#~jT@gx&1(X@rgx&1$@gK7]@rrrNrrArArArBr#Grr#c @r)r$zJhttps://download.pytorch.org/models/efficientnet_b6_lukemelas-24a108a5.pthirirgnU@gv:X@rg rh3@g$d@rrrNrrArArArBr$_rr$c @r)r%zJhttps://download.pytorch.org/models/efficientnet_b7_lukemelas-c5b4e57e.pthiXricrg+U@g'1:X@rgsh|B@go@rrrNrrArArArBr%wrr%c @r)r&zBhttps://download.pytorch.org/models/efficientnet_v2_s-dd5fe13b.pthrri8nGrg;OU@gx&18X@rgZd @gVT@rrrN rErFrGrrrrr_COMMON_META_V2rrrArArArBr&0r&c @r)r'zBhttps://download.pytorch.org/models/efficientnet_v2_m-dc08266a.pthri:rgI +GU@gDlIX@rgE8@gQj@rrrNrrArArArBr'rr'c @sNeZdZedeeddejdddieddddd id d d d dZ e Z dS)r(zBhttps://download.pytorch.org/models/efficientnet_v2_l-59c71312.pthr)?rr)rrrmeanstdiHfrgʡEsU@gOnrX@rg ףp= L@gI +i|@rrrN) rErFrGrrrrrrrrrArArArBr(s4r( pretrained)rT)rrcKs<t|}tdddd\}}t||dd|||fi|S)aEfficientNet B0 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B0_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B0_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B0_Weights :members: r)rNrr~r{)rverifyrrrrrrr}rrArArBr) r)cK<t|}tdddd\}}t||dd|||fi|S)aEfficientNet B1 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B1_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B1_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B1_Weights :members: r*rN皙?rr~r{)rr rrrr rArArBr*r r*cKr )aEfficientNet B2 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B2_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B2_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B2_Weights :members: r+r 333333?rr~333333?)r r rrrr rArArBr+&r r+cKr )aEfficientNet B3 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B3_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B3_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B3_Weights :members: r,rffffff?rr~r)r!r rrrr rArArBr,E  r,cKr )aEfficientNet B4 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B4_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B4_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B4_Weights :members: r-r?rr~皙?)r"r rrrr rArArBr-irr-cKLt|}tdddd\}}t||dd|||fdttjdd d i|S) aEfficientNet B5 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B5_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B5_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B5_Weights :members: r.g?g@rr~rr^MbP?{Gz?epsmomentum)r#r rrrrr rr rArArBr.  r.cKr) aEfficientNet B6 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B6_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B6_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B6_Weights :members: r/rg@rr~rr^rrr)r$r rrrrr rr rArArBr/rr/cKr) aEfficientNet B7 model architecture from the `EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks `_ paper. Args: weights (:class:`~torchvision.models.EfficientNet_B7_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_B7_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_B7_Weights :members: r0g@g@rr~rr^rrr)r%r rrrrr rr rArArBr0rr0cKDt|}td\}}t||dd|||fdttjddi|S)a Constructs an EfficientNetV2-S architecture from `EfficientNetV2: Smaller Models and Faster Training `_. Args: weights (:class:`~torchvision.models.EfficientNet_V2_S_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_V2_S_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_V2_S_Weights :members: r1r~r{r^rr)r&r rrrrr rr rArArBr1    r1cKr)a Constructs an EfficientNetV2-M architecture from `EfficientNetV2: Smaller Models and Faster Training `_. Args: weights (:class:`~torchvision.models.EfficientNet_V2_M_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_V2_M_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_V2_M_Weights :members: r2r~rr^rr)r'r rrrrr rr rArArBr2"rr2cKr)a Constructs an EfficientNetV2-L architecture from `EfficientNetV2: Smaller Models and Faster Training `_. Args: weights (:class:`~torchvision.models.EfficientNet_V2_L_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.EfficientNet_V2_L_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.efficientnet.EfficientNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.EfficientNet_V2_L_Weights :members: r3r~rr^rr)r(r rrrrr rr rArArBr3Hrr3)OrrW dataclassesr functoolsrtypingrrrrrr r r rr r torchvision.opsrops.miscrrtransforms._presetsrrutilsr_apirrr_metar_utilsrrr__all__r4rMrZrKrQr[rrHrJboolrstrrrrIrrrrr r!r"r#r$r%r&r'r(rr)r*r+r,r-r.r/r0r1r2r3rArArArBs   (   C=r  80""###$$