o h; @slddlmZddlmZddlmZmZmZddlZddlm Z m Z ddl m Z ddlmZmZdd lmZdd lmZdd lmZd d lmZmZmZd dlmZd dlmZmZgdZ Gddde j!Z"Gddde j#Z$GdddZ%Gddde j#Z&de'e%de(deede)dede&f dd Z*d!ed"d#d$Z+Gd%d&d&eZ,Gd'd(d(eZ-Gd)d*d*eZ.Gd+d,d,eZ/eed-e,j0fd.dd/d0dee,de)dede&fd1d2Z1eed-e-j0fd.dd/d0dee-de)dede&fd3d4Z2eed-e.j0fd.dd/d0dee.de)dede&fd5d6Z3eed-e/j0fd.dd/d0dee/de)dede&fd7d8Z4dS)9)Sequence)partial)AnyCallableOptionalN)nnTensor) functional)Conv2dNormActivationPermute)StochasticDepth)ImageClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) ConvNeXtConvNeXt_Tiny_WeightsConvNeXt_Small_WeightsConvNeXt_Base_WeightsConvNeXt_Large_Weights convnext_tinyconvnext_small convnext_baseconvnext_largec@seZdZdedefddZdS) LayerNorm2dxreturncCs>|dddd}t||j|j|j|j}|dddd}|S)Nrr r)ZpermuteFZ layer_normZnormalized_shapeweightbiasepsselfr!r*a/home/www/facesmatcher.com/frenv_anti/lib/python3.10/site-packages/torchvision/models/convnext.pyforward szLayerNorm2d.forwardN)__name__ __module__ __qualname__rr,r*r*r*r+r sr c sReZdZ d dededeedejfddffdd Zd e de fd d Z Z S) CNBlockN layer_scalestochastic_depth_prob norm_layer.r"c st|durttjdd}ttj||dd|ddtgd||tj|d|dd t tjd||dd tgd |_ t t |d d ||_t|d |_dS) Nư>r'r#T) kernel_sizepaddinggroupsr&)rr r#r)Z in_featuresZ out_featuresr&)rr#rr rrow)super__init__rr LayerNorm SequentialConv2dr LinearZGELUblock ParametertorchZonesr1r stochastic_depth)r)dimr1r2r3 __class__r*r+r=(s    zCNBlock.__init__inputcCs&|j||}||}||7}|SN)r1rBrE)r)rIresultr*r*r+r,?s zCNBlock.forwardrJ) r-r.r/floatrrrModuler=rr, __classcell__r*r*rGr+r0'sr0c@s8eZdZdedeededdfddZdefdd ZdS) CNBlockConfiginput_channels out_channels num_layersr"NcCs||_||_||_dSrJ)rPrQrR)r)rPrQrRr*r*r+r=Hs zCNBlockConfig.__init__cCs>|jjd}|d7}|d7}|d7}|d7}|jdi|jS)N(zinput_channels={input_channels}z, out_channels={out_channels}z, num_layers={num_layers})r*)rHr-format__dict__)r)sr*r*r+__repr__Rs zCNBlockConfig.__repr__)r-r.r/intrr=strrXr*r*r*r+rOFs  rOcseZdZ     ddeedededed eed e j fd eed e j fd e d dffdd Z de d e fddZde d e fddZZS)rr4N block_settingr2r1 num_classesrB.r3kwargsr"c stt||stdt|trtdd|Ds!td|dur't}|dur1t t dd}g}|dj } | t d| d d d|dd d td d |D} d} |D]D} g} t| jD]}|| | d}| || j ||| d7} q]| tj| | jdur| t|| j tj| j | jdddqTtj||_td|_|d}|jdur|jn|j }t||tdt|||_|D] }t|tjtjfrtjj|jdd|jdurtj |jqdS)Nz%The block_setting should not be emptycSsg|]}t|tqSr*) isinstancerO).0rWr*r*r+ ksz%ConvNeXt.__init__..z/The block_setting should be List[CNBlockConfig]r4r5rr#r:T)r7strider8r3Zactivation_layerr&css|]}|jVqdSrJ)rR)racnfr*r*r+ sz$ConvNeXt.__init__..g?rr )r7rcg{Gz?)Zstd)!r<r=r ValueErrorr`rall TypeErrorr0rr rPappendr sumrangerRrr?rQr@featuresZAdaptiveAvgPool2davgpoolZFlattenrA classifiermodulesinitZ trunc_normal_r%r&Zzeros_)r)r]r2r1r^rBr3r_ZlayersZfirstconv_output_channelsZtotal_stage_blocksZstage_block_idrdZstage_Zsd_probZ lastblockZlastconv_output_channelsmrGr*r+r=\sp          zConvNeXt.__init__r!cCs"||}||}||}|SrJ)rmrnror(r*r*r+ _forward_impls   zConvNeXt._forward_implcCs ||SrJ)rtr(r*r*r+r,s zConvNeXt.forward)r[r4r\NN)r-r.r/listrOrLrYrrrrMrr=rrtr,rNr*r*rGr+r[s2 Nrr]r2weightsprogressr_r"cKsR|durt|dt|jdt|fd|i|}|dur'||j|dd|S)Nr^ categoriesr2T)rwZ check_hash)rlenmetarZload_state_dictZget_state_dict)r]r2rvrwr_modelr*r*r+ _convnexts r|) r}zNhttps://github.com/pytorch/vision/tree/main/references/classification#convnexta These weights improve upon the results of the original paper by using a modified version of TorchVision's `new training recipe `_. )Zmin_sizerxZrecipeZ_docsc @DeZdZedeedddieddddd id d d d ZeZdS)rz>https://download.pytorch.org/models/convnext_tiny-983f1562.pthZ crop_sizeZ resize_sizeiH< ImageNet-1KgzGT@gMbX X@zacc@1zacc@5gm@gV-G[@Z num_paramsZ_metricsZ_ops _file_sizeurlZ transformsrzN r-r.r/rrr _COMMON_META IMAGENET1K_V1DEFAULTr*r*r*r+r$ rc @r~)rz?https://download.pytorch.org/models/convnext_small-0c510722.pthrriHZrgClT@g)X@rg|?5^!@g"~g@rrNrr*r*r*r+rrrc @r~)rz>https://download.pytorch.org/models/convnext_base-6075fbad.pthrrihGrgU@gHz7X@rg(\µ.@g/$!u@rrNrr*r*r*r+rrrc @r~)rz?https://download.pytorch.org/models/convnext_large-ea097f82.pthrrri rg"~U@gX9v>X@rg|?5.A@gK@rrNrr*r*r*r+rrrZ pretrained)rvT)rvrwcKXt|}tdddtdddtdddtdddg}|dd }t||||fi|S) aConvNeXt Tiny model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Tiny_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Tiny_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.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Tiny_Weights :members: `r# Nr2g?)rverifyrOpopr|rvrwr_r]r2r*r*r+r"      rcKr) aConvNeXt Small model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Small_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Small_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.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Small_Weights :members: rrr#rrNr2g?)rrrOrr|rr*r*r+rA      rcKr) aConvNeXt Base model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Base_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Base_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.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Base_Weights :members: r#iirNr2?)rrrOrr|rr*r*r+rbrrcKr) aConvNeXt Large model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Large_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Large_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.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Large_Weights :members: rrr#rirNr2r)rrrOrr|rr*r*r+rrr)5collections.abcr functoolsrtypingrrrrDrrZtorch.nnr r$Zops.miscr r Zops.stochastic_depthr Ztransforms._presetsrutilsrZ_apirrr_metarZ_utilsrr__all__r>r rMr0rOrrurLboolr|rrrrrrrrrrr*r*r*r+s       Y  **