kh; ddlmZddlmZmZmZmZmZddlZddlm Z m Z ddl m Z ddlmZmZddlmZdd lmZdd lmZd d lmZmZmZd d lmZd dlmZmZgdZ Gdde jBZ"Gdde jFZ$GddZ%Gdde jFZ&dee%de'deede(dede&f dZ)ded d!d"Z*Gd#d$eZ+Gd%d&eZ,Gd'd(eZ-Gd)d*eZ.eed+e+j^f,dd-d.dee+de(dede&fd/Z0eed+e,j^f,dd-d.dee,de(dede&fd0Z1eed+e-j^f,dd-d.dee-de(dede&fd1Z2eed+e.j^f,dd-d.dee.de(dede&fd2Z3y)3)partial)AnyCallableListOptionalSequenceN)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_largeceZdZdedefdZy) LayerNorm2dxreturnc|jdddd}tj||j|j|j |j }|jdddd}|S)Nrr r)permuteF layer_normnormalized_shapeweightbiasepsselfr$s W/var/www/teggl/fontify/venv/lib/python3.12/site-packages/torchvision/models/convnext.pyforwardzLayerNorm2d.forwardsW IIaAq ! LLD114;; 488 T IIaAq !N)__name__ __module__ __qualname__r r2r3r1r#r#sFr3r#c heZdZ d dededeedejfddffd Zde de fd Z xZ S) CNBlockN layer_scalestochastic_depth_prob norm_layer.r%c t||ttjd}tj tj ||dd|dtgd||tj|d|zd tjtjd|z|d tgd |_ tjtj|d d |z|_t|d |_y) Nư>r.r'T) kernel_sizepaddinggroupsr-)rr r'r) in_features out_featuresr-)rr'rr rrow)super__init__rr LayerNorm SequentialConv2drLinearGELUblock Parametertorchonesr:rstochastic_depth)r0dimr:r;r< __class__s r1rIzCNBlock.__init__'s    48J]] IIc3Aq4 P L ! sO II#AG$ G GGI II!c'$ G L !  << 31(= (KL /0Eu Mr3inputcn|j|j|z}|j|}||z }|SN)r:rOrS)r0rVresults r1r2zCNBlock.forward>s9!!DJJu$55&&v.% r3rX) r4r5r6floatrrr ModulerIr r2 __classcell__rUs@r1r9r9&s_ :> NN % N Xc299n56 N  N.Vr3r9c6eZdZdedeededdfdZdefdZy) CNBlockConfiginput_channels out_channels num_layersr%Nc.||_||_||_yrX)r`rarb)r0r`rarbs r1rIzCNBlockConfig.__init__Gs -($r3c|jjdz}|dz }|dz }|dz }|dz }|jdi|jS)N(zinput_channels={input_channels}z, out_channels={out_channels}z, num_layers={num_layers})r7)rUr4format__dict__)r0ss r1__repr__zCNBlockConfig.__repr__QsX NN # #c ) .. ,, (( Sqxx($--((r3)r4r5r6intrrIstrrjr7r3r1r_r_Es=%%sm% %  %)#)r3r_ceZdZ ddeededededeede jfdeede jfd e d dffd Z d e d e fd Zd e d e fdZxZS)rN block_settingr;r: num_classesrO.r<kwargsr%c t|t||s tdt |t r't |Dcgc]}t |tc}s td|t}|ttd}g} |dj} | jtd| ddd|dd td |D} d} |D]} g}t!| j"D]5}|| z| d z z }|j|| j||| d z } 7| jt%j&|| j(| jt%j&|| jt%j*| j| j(d d t%j&| |_t%j.d |_|d}|j( |j(n |j}t%j&||t%j2d t%j4|||_|j9D]}t |t$j*t$j4fs.t$j:j=|j>d|j@ft$j:jC|j@ycc}w)Nz%The block_setting should not be emptyz/The block_setting should be List[CNBlockConfig]r>r?rr'rDT)rAstriderBr<activation_layerr-c34K|]}|jywrX)rb).0cnfs r1 z$ConvNeXt.__init__..s IC Isg?rr )rArrg{Gz?)std)"rHrIr ValueError isinstancerallr_ TypeErrorr9rr#r`appendr sumrangerbr rKrarLfeaturesAdaptiveAvgPool2davgpoolFlattenrM classifiermodulesinit trunc_normal_r,r-zeros_)r0rnr;r:rorOr<rprilayersfirstconv_output_channelstotal_stage_blocksstage_block_idrvstage_sd_prob lastblocklastconv_output_channelsmrUs r1rIzConvNeXt.__init__[s D!DE E]H5#er>s`az!]?[>s:tMN N =E   $7J"$%2!$4$C$C! )%!%  ! I= II  C%'E3>>* $/.@DVY\D\] U3#5#5{GLM!#  $ MM"--/ 0+ MM"3#5#56 #"4"4c6F6FTU^_` $ v. ++A. !"% &/&<&<&HI " "iNfNf !-- / 0"**Q-KcepAq  +A!bii34%%ahhD%966%GGNN166*  +s?tsK5r$cl|j|}|j|}|j|}|SrX)rrrr/s r1 _forward_implzConvNeXt._forward_impls0 MM!  LLO OOA r3c$|j|SrX)rr/s r1r2zConvNeXt.forwards!!!$$r3)gr>iNN)r4r5r6rr_rZrkrrr r[rrIr rr2r\r]s@r1rrZs(+!489=L+M*L+ %L+ L+  L+ bii01 L+Xc299n56L+L+ L+\v& %%F%r3rrnr;weightsprogressrpr%c |#t|dt|jdt|fd|i|}|"|j |j |d|S)Nro categoriesr;T)r check_hash)rlenmetarload_state_dictget_state_dict)rnr;rrrpmodels r1 _convnextrsdfmSl9S5TU ] Z:O ZSY ZE g44hSW4XY Lr3) rzNhttps://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 `_. )min_sizerrecipe_docsc TeZdZedeeddieddddd id d d  ZeZy)rz>https://download.pytorch.org/models/convnext_tiny-983f1562.pth crop_size resize_sizeiH< ImageNet-1KgzGT@gMbX X@zacc@1zacc@5gm@gV-G[@ num_params_metrics_ops _file_sizeurl transformsrN r4r5r6rrr _COMMON_META IMAGENET1K_V1DEFAULTr7r3r1rrsS L.#3O    "##  !  M Gr3rc TeZdZedeeddieddddd id d d  ZeZy)rz?https://download.pytorch.org/models/convnext_small-0c510722.pthrriHZrgClT@g)X@rg|?5^!@g"~g@rrNrr7r3r1rrsS M.#3O    "##  !  M Gr3rc TeZdZedeeddieddddd id d d  ZeZy)rz>https://download.pytorch.org/models/convnext_base-6075fbad.pthrrihGrgU@gHz7X@rg(\µ.@g/$!u@rrNrr7r3r1rrsS L.#3O    "##  !  M Gr3rc TeZdZedeeddieddddd id d d  ZeZy)rz?https://download.pytorch.org/models/convnext_large-ea097f82.pthrrri rg"~U@gX9v>X@rg|?5.A@gK@rrNrr7r3r1rr sS M.#3O    ###  !  M Gr3r pretrained)rT)rrc tj|}tdddtdddtdddtdddg}|jdd }t ||||fi|S) aConvNeXt Tiny model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Tiny_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Tiny_Weights` below for more details and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Tiny_Weights :members: `r' Nr;g?)rverifyr_poprrrrprnr;s r1rr!sy&$**73G b#q!c3"c3"c4# M #JJ'>D ]$97H WPV WWr3c tj|}tdddtdddtdddtdddg}|jdd }t ||||fi|S) 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'rrNr;g?)rrr_rrrs r1rr@sy*%++G4G b#q!c3"c3#c4# M #JJ'>D ]$97H WPV WWr3c tj|}tdddtdddtdddtdddg}|jdd }t ||||fi|S) aConvNeXt Base model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Base_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Base_Weights` below for more details and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Base_Weights :members: r'iirNr;?)rrr_rrrs r1r r asy&$**73G c3"c3"c4$dD!$ M #JJ'>D ]$97H WPV WWr3c tj|}tdddtdddtdddtdddg}|jdd }t ||||fi|S) aConvNeXt Large model architecture from the `A ConvNet for the 2020s `_ paper. Args: weights (:class:`~torchvision.models.convnext.ConvNeXt_Large_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.convnext.ConvNeXt_Large_Weights` below for more details and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.convnext.ConvNext`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ConvNeXt_Large_Weights :members: rrr'rirNr;r)rrr_rrrs r1r!r!sy*%++G4G c3"c3"c4$dD!$ M #JJ'>D ]$97H WPV WWr3)4 functoolsrtypingrrrrrrQr r torch.nnr r)ops.miscr rops.stochastic_depthrtransforms._presetsrutilsr_apirrr_metar_utilsrr__all__rJr#r[r9r_rrZboolrrrrrrrrrr r!r7r3r1rs:: $425'66'B ",,bii>))*V%ryyV%r & k "     &&^  K([(K([(,0E0S0S!TU@DW[Xh'<=XPTXgjXowXVX:,0F0T0T!UV37$X/0XCGXZ]X XWX>,0E0S0S!TU@DW[Xh'<=XPTXgjXowXVX:,0F0T0T!UV37$X/0XCGXZ]X XWXr3