RЦiSrSSKJr SSKJr SSKJrJrJrJ r J r J r J r SSK r SSKJr SSKJrJr SSKJrJrJrJrJrJrJrJrJrJrJrJrJ r J!r! SS K"J#r# SS K$J%r% SS K&J'r'J(r(J)r) SS K*J+r+J,r,J-r- S /r."SS\R^5r0"SS\R^5r1"SS\R^5r2"SS\R^5r3"SS\R^5r4"SS\R^5r5S\6S\74Sjr8SSS\\"\S S!9SS4S"\9S#\9S\6S$\7S%\S&\S\Rt4S'jjr;"S(S \R^5rS?S@S:SSA9_SB\B"S9SCSDS@SSE9_SF\B"S9SCSDS@SSE9_SG\B"S9SCSDS@SSE9_SH\B"S9SCSDS@SSE9_SI\B"S9SCSDS@SSE9_SJ\B"S9SKSLS@SSE9_SM\B"S9SNSSO9_SP\B"S9SNSSO9_SQ\B"S9SNSSO9_SR\B"S9SNSSO9_SS\B"S9SNSSO9_ST\B"S9SNSSO9_SU\B"S9S:SVSWS@SX9_SY\B"S9S:SVSWS@SZS[9_\B"S9S:SVSWS@SX9\B"S9S:SVSWS@SZS[9\B"S9S@S>S?SCS:SZS\9\B"S9S:S]SWS@SX9\B"S:SZS^9\B"S:SZS^9\B"S9S:S]SWS@SX9\B"S:SZS^9\B"S:S_9\B"S:SZS^9\B"S9S:SZS]SWS@S`9\B"S9S:SZS]SWS@S`9\B"S:SZS^9Sa. E5rC\,SyS2\7S3\S\<4Sbjj5rD\,SyS2\7S3\S\<4Scjj5rE\,SyS2\7S3\S\<4Sdjj5rF\,SyS2\7S3\S\<4Sejj5rG\,SyS2\7S3\S\<4Sfjj5rH\,SyS2\7S3\S\<4Sgjj5rI\,SyS2\7S3\S\<4Shjj5rJ\,SyS2\7S3\S\<4Sijj5rK\,SyS2\7S3\S\<4Sjjj5rL\,SyS2\7S3\S\<4Skjj5rM\,SyS2\7S3\S\<4Sljj5rN\,SyS2\7S3\S\<4Smjj5rO\,SyS2\7S3\S\<4Snjj5rP\,SyS2\7S3\S\<4Sojj5rQ\,SyS2\7S3\S\<4Spjj5rR\,SyS2\7S3\S\<4Sqjj5rS\,SyS2\7S3\S\<4Srjj5rT\,SyS2\7S3\S\<4Ssjj5rU\,SyS2\7S3\S\<4Stjj5rV\,SyS2\7S3\S\<4Sujj5rW\-"\XSBSFSGSHSISJSMSPSQSRSSSTS8SXS.n[TU]5 U=(d UnU =(d [n U =(d [[SS9n U=(d Un[ X#-5nU b'US:wd Xe:wdX:waU "UU4UUUSU U S.UD6UlOSUlU "U40UD6UlU "UUS4XFUS .UD6UlU "U40UD6Ul U "UUS4XWS .UD6Ul U S :a[U 5Ulg[R"5Ulg) aInitialize PreActBasic block. Args: in_chs: Input channels. out_chs: Output channels. bottle_ratio: Bottleneck ratio (not used in basic block). stride: Stride for convolution. dilation: Dilation rate. first_dilation: First dilation rate. groups: Group convolution size. act_layer: Activation layer type. conv_layer: Convolution layer type. norm_layer: Normalization layer type. proj_layer: Projection/downsampling layer type. drop_path_rate: Stochastic depth drop rate. devicedtype  num_groupsNrTr-r.r/preactr2r3r-r.r0)r.r0r)super__init__rrrr downsamplenorm1conv1norm2conv2rnnIdentity drop_pathselfr*r+r,r-r.r/r0r1r2r3r4r5r8r9ddmid_chs __class__s S/var/www/html/ai-image-ml/venv/lib/python3.13/site-packages/timm/models/resnetv2.pyrBPreActBasic.__init__5s$B / '38,9 G7XS.n[TU]5 U=(d UnU =(d [n U =(d [[SS9n U=(d Un[ X#-5nU bU "UU4UUUSU U S.UD6UlOSUlU "U40UD6UlU "UUS40UD6UlU "U40UD6Ul U "UUS4XFUS .UD6Ul U "U40UD6Ul U "UUS40UD6Ul U S :a[U 5Ulg[R"5Ulg) aInitialize PreActBottleneck block. Args: in_chs: Input channels. out_chs: Output channels. bottle_ratio: Bottleneck ratio. stride: Stride for convolution. dilation: Dilation rate. first_dilation: First dilation rate. groups: Group convolution size. act_layer: Activation layer type. conv_layer: Convolution layer type. norm_layer: Normalization layer type. proj_layer: Projection/downsampling layer type. drop_path_rate: Stochastic depth drop rate. r7r:r;NTr=rr?r@r)rArBrrrrrCrDrErFrGnorm3conv3rrHrIrJrKs rPrBPreActBottleneck.__init__s3B / '38,9 G7[>[c]>[ >[  >[  >[%SM>[>[ )>[!*>[!*>[!*>[">[>[@*%,,rRruc^\rSrSrSrSS\S\\S\S\S\S\\S \S \\S \\S \\S \\S\4U4Sjjjr SSjr S\ RS\ R4Sjr SrU=r$) BottleneckzUNon Pre-activation bottleneck block, equiv to V1.5/V1b Bottleneck. Used for ViT. r*r+r,r-r.r/r0r1r2r3r4r5c X>XS.n[TU]5 U=(d UnU=(d [RnU =(d [n U =(d [ [ SS9n U=(d Un[X#-5nU bU "UU4UUSU U S.UD6UlOSUlU "UUS40UD6Ul U "U40UD6Ul U "UUS4XFUS.UD6Ul U "U40UD6Ul U "UUS40UD6Ul U "U4S S0UD6UlU S :a [U 5O[R "5UlU"S S 9Ulg) Nr7r:r;F)r-r.r>r2r3rr?r@ apply_actrT)inplace)rArBrHReLUrrrrrCrErDrGrFryrxrrIrJact3rKs rPrBBottleneck.__init__sJ" / '38( ,9 G7r2r3c >XS.n [T U]5 U"XS4SU0U D6UlU(a[R"5UlgU"U4SS0U D6Ulg)Nr7rr-rF)rArBconvrHrInorm) rLr*r+r-r.r/r>r2r3r8r9rMrOs rPrBDownsampleConv.__init__IsU / vG&GBG %+BKKM G1[u1[XZ1[ rRr\rScBURURU55$zLForward pass. Args: x: Input tensor. Returns: Downsampled tensor. )rrrLr\s rPrbDownsampleConv.forward[syy1&&rR)rrrrNTNNNNrhrirjrkrlrmr boolrrBrorprbrqrrrss@rPrrFs. ,0-1-1\\\ \  \ %SM \\!*\!*\\$ ' '%,, ' 'rRrc^\rSrSrSrSS\S\S\S\S\\S\S \\S \\4U4S jjjr S \ RS \ R4Sjr Sr U=r$) DownsampleAvgigz/AvgPool downsampling as in 'D' ResNet variants.r*r+r-r.r/r>r2r3c r>XS.n [TU]5 US:XaUOSn US:dUS:a1U S:Xa US:a[O[Rn U "SU SSS9UlO[R "5UlU"XS4SS0U D6UlU(a[R "5UlgU"U4SS0U D6Ulg) Nr7rTF) ceil_modecount_include_padr-r) rArBrrH AvgPool2dpoolrIrr)rLr*r+r-r.r/r>r2r3r8r9rM avg_stride avg_pool_fnrOs rPrBDownsampleAvg.__init__js / '1}V! A:A+5?x!|-QSQ]Q]K#AzTUZ[DI DIvB!BrB %+BKKM G1[u1[XZ1[ rRr\rSc`URURURU555$r)rrrrs rPrbDownsampleAvg.forwards$yy499Q<011rR)rrrrrrss@rPrrgs9 ,0-1-1\\\ \  \ %SM \\!*\!*\\0 2 2%,, 2 2rRrc^\rSrSrSrSSSS\SSS4S\S\S \S \S \S \S \S\S\ \ \S\ S\ \ S\ \ S\ \ S\ 4U4Sjjjr S\RS\R4SjrSrU=r$) ResNetStageiz ResNet Stage.rrFNr*r+r-r.depthr,r0avg_down block_dprblock_fnr1r2r3 block_kwargsc >[TU]5 SUlUS;aSOSn[XU S9nU(a[O[ nUn[ R"5Ul[U5HXnU (aU UOSnUS:XaUOSnURR[U5U "UU4UUUUUUUS.UDUD65 UnUnSnMZ g) NF)rrrr)r1r2r3rer)r-r.r,r0r/r4r5) rArBgrad_checkpointingdictrrrH Sequentialblocksrange add_modulestr)rLr*r+r-r.rr,r0rrrr1r2r3rr/ layer_kwargsr4prev_chs block_idxr5rOs rPrBResNetStage.__init__s" "'&&0aiS]^ &.]N mmo uI5>Yy1BN(A~V1F KK " "3y>8 4!)-%- 4 4 4 H%NJ%&rRr\rScUR(a;[RR5(d[ UR U5nU$UR U5nU$)zgForward pass through all blocks in the stage. Args: x: Input tensor. Returns: Output tensor. )rrojit is_scriptingr rrs rPrbResNetStage.forwardsJ  " "599+A+A+C+Ct{{A.A AArR)rr)rhrirjrkrlrurmrnrr rrrrBrorprbrqrrrss@rPrrs#'"/3!1,0-1-1+++ +  +  + +++ U ,++ )+!*+!*+ ++Z  %,,  rRr stem_typerScH[SVs/sHoU;PM sn5$s snf)zCheck if stem type is deep (has multiple convolutions). Args: stem_type: Type of stem to check. Returns: True if stem is deep, False otherwise. )deeptiered)any)rss rP is_stem_deeprs& (:;(:1Y(:; <<;s@Tr:r;r*r+r>r2r3cXgS.n[5n US;de[U5(aSU;aSU-S-US-4n O US-US-4n U"X S4SSS.UD6U S 'U"U S40UD6U S 'U"U SU S 4SS S.UD6U S 'U"U S 40UD6U S 'U"U S U4SS S.UD6U S'U(d U"U40UD6U S'O"U"X4SSS.UD6U S'U(d U"U40UD6U S'SU;a4[R"S S5U S'[R"SSSS9U S'O/SU;a[ SSSSS9U S'O[R"SSS S9U S'[R "U 5$)Nr7)rfixedsamer deep_fixed deep_samerrr?rr) kernel_sizer-rErDrrGrFryrxrrrrepad)rr-paddingrrmax)rrrH ConstantPad2d MaxPool2drr) r*r+rr>r2r3r8r9rMstemstem_chss rPcreate_resnetv2_stemrs +B =D Z ZZ ZI y G q('Q,7H1 gl3H"6A;VAaVSUVW "8A;5"5W "8A; [ST[XZ[W "8A;5"5W "8A;WQqWTVWW &w5"5DM"&QqQbQV %g44DL)&&q"-U ||!QGV 9 $U!VTV ||!QGV == rRc' ^\rSrSrSrSSSSSSS S S S S S \R \"\SS9\ SSS SS4S\ \ S\ \ S4S\ S\ S\ S\ S\ S\ S\ S\S\S\S\S\S \S!\S"\S#\S$\4&U4S%jjjr\R&R(S@S$\S&S4S'jj5r\R&R)5SAS(\ S)\ S&S4S*jj5r\R&R(SBS+\S&\\ \44S,jj5r\R&R(S@S-\S&S4S.jj5r\R&R(S&\R64S/j5rSCS\ S\\ S&S4S0jjrSDS1\R>S2\\ \ \ \ 4S3\S4\S5\ S6\S&\ \ \R>\ \R>\ \R>444S7jjr!SES2\ \ \ \ 4S8\S9\4S:jjr"S1\R>S&\R>4S;jr#SBS1\R>S<\S&\R>4S=jjr$S1\R>S&\R>4S>jr%S?r&U=r'$)Fr&i z7Implementation of Pre-activation (v2) ResNet mode. )iir?avgr:rrrFTrr;reNlayerschannels. num_classesin_chans global_pool output_stride width_factorrrrr>basicr,r1r3r2 drop_rater5rYcT>[T%U]5 UUS.nX0lX@lUUlUn[ XS9n[ U5n/Ul[UU-5n[UUU U 4UUS.UD6Ul U (a[U 5(aSOSOSnURR[USUS95 UnS nS n[UUS S 9nU (aU (a[O[ nOU (ae["n[$R&"5Ul[+[-XU55Hununn n![U U-5n"US :XaS OSn#UU:aUU#-nS n#[/UU"4U#UUU U UUUU!US. UD6n$U"nUU#-nU=R[UUSU3S9/- slUR(R1[3U5U$5 M U=UlUlU (aU"UR440UD6O[$R8"5Ul[=UR4U4UURS S.UD6UlURAUS9 g)a: Args: layers (List[int]) : number of layers in each block channels (List[int]) : number of channels in each block: num_classes (int): number of classification classes (default 1000) in_chans (int): number of input (color) channels. (default 3) global_pool (str): Global pooling type. One of 'avg', 'max', 'avgmax', 'catavgmax' (default 'avg') output_stride (int): output stride of the network, 32, 16, or 8. (default 32) width_factor (int): channel (width) multiplication factor stem_chs (int): stem width (default: 64) stem_type (str): stem type (default: '' == 7x7) avg_down (bool): average pooling in residual downsampling (default: False) preact (bool): pre-activation (default: True) act_layer (Union[str, nn.Module]): activation layer norm_layer (Union[str, nn.Module]): normalization layer conv_layer (nn.Module): convolution module drop_rate: classifier dropout rate (default: 0.) drop_path_rate: stochastic depth rate (default: 0.) zero_init_last: zero-init last weight in residual path (default: False) r7)r1)r2r3z stem.conv3 stem.convz stem.normr)num_chs reductionmodulerT) stagewiser) r-r.rr,rr1r2r3rrzstages.) pool_typeruse_convrYN)!rArBrrrrr feature_inforrrrappendrrr(rurrHrstages enumerateziprrr num_featureshead_hidden_sizerIrrhead init_weights)&rLrrrrrrrrrrr>rr,r1r3r2rr5rYr8r9rMwf stem_featr curr_strider. block_dprsr stage_idxdcbdprr+r-stagerOs& rPrBResNetV2.__init__ sfX  /& " ' H !), !(R-0(      "!    SY\)%<%<\+^i    h!I!VW .~vQUV &+{1AH 9!Hmmo '0Vz1R'S #I|1d$QV,G#q.QaFm+F"!)!#%%!EH 6 !K   $x;Y`aj`kWl"m!n n  KK " "3y>5 91(T45=<D1;AJt007B7r{{} "     "nn      8rRrSc4[[[US9U5 g)zInitialize model weights.rN)r!r _init_weights)rLrYs rPrResNetV2.init_weightss GM.I4PrRcheckpoint_pathprefixc[XU5 g)zLoad pretrained weights.N) _load_weights)rLrrs rPload_pretrainedResNetV2.load_pretraineds dV4rRcoarsec0[SU(aSOSS/S9nU$)z"Group parameters for optimization.z^stemz^stages\.(\d+))z^stages\.(\d+)\.blocks\.(\d+)N)z^norm)i)rr)r)rLr matchers rP group_matcherResNetV2.group_matchers)(.$8$5 rRenablec6URH nXlM g)z)Enable or disable gradient checkpointing.N)rr)rLrrs rPset_grad_checkpointingResNetV2.set_grad_checkpointingsA#) rRc.URR$)zGet the classifier head.)rfcrXs rPget_classifierResNetV2.get_classifiersyy||rRcFXlURRX5 g)zReset the classifier head. Args: num_classes: Number of classes for new classifier. global_pool: Global pooling type. N)rrreset)rLrrs rPreset_classifierResNetV2.reset_classifiers'  1rRr\indicesr stop_early output_fmtintermediates_onlycUS;dS5e/n[SU5upSn URSSupURH*n U "U5nURSSU S-U S-4:XdM(UnM, X;aURW5 [ UR 5n[ RR5(dU(d UR nOUR SU n[USS 9HXun nU"U5nX;dMX:Xa-U(aURU5OUnURU5 MGURU5 MZ U(aU$X:XaURU5nX4$) aForward features that returns intermediates. Args: x: Input image tensor indices: Take last n blocks if int, all if None, select matching indices if sequence norm: Apply norm layer to compatible intermediates stop_early: Stop iterating over blocks when last desired intermediate hit output_fmt: Shape of intermediate feature outputs intermediates_only: Only return intermediate features Returns: )NCHWzOutput shape must be NCHW.rNrr)start) rshaperrlenrrorrrr)rLr\rrrr r! intermediates take_indices max_indexfeat_idxHWrx_downlast_idxrrx_inters rPforward_intermediatesResNetV2.forward_intermediatessN*Y&D(DD& "6q'"B wwrs|IIDQAwwrs|Aq1u~-  #   (t{{# 99 ! ! # #:[[F[[),F(q9OHeaA''.2diilG!((1!((+ :    ! ArR prune_norm prune_headc[SU5upEURSUUlU(a[R"5UlU(aUR SS5 U$)z?Prune layers not required for specified intermediates. r$Nrr)rrrHrIrr)rLrr4r5r*r+s rPprune_intermediate_layers"ResNetV2.prune_intermediate_layerssN#7q'"B kk*9-  DI   ! !!R (rRclURU5nURU5nURU5nU$)zjForward pass through feature extraction layers. Args: x: Input tensor. Returns: Feature tensor. )rrrrs rPforward_featuresResNetV2.forward_featuress0 IIaL KKN IIaLrR pre_logitscPU(aURXS9$URU5$)zForward pass through classifier head. Args: x: Input features. pre_logits: Return features before final linear layer. Returns: Classification logits or features. )r<)r)rLr\r<s rP forward_headResNetV2.forward_heads&7Atyyy2RdiiPQlRrRcJURU5nURU5nU$)zGForward pass. Args: x: Input tensor. Returns: Output logits. )r:r>rs rPrbResNetV2.forwards)  ! !! $   a rR) rrrrrrrrrr)Tzresnet/F)N)NFFr#F)rFT)(rhrirjrkrlrHrrrrrrmr rrrnrrBrorignorerr rrrrModulerr rrpr r2r7r:r>rbrqrrrss@rPr&r& s} )?#$!# !""&"$''#**  %,, 5<<  Sell S S S  %,,  rRrnamerYc[U[R5(d%SU;ar[U[R5(aS[RR UR SSS9 [RRUR5 g[U[R5(aa[RRUR SSS9 URb*[RRUR5 gg[U[R[R[R45(aS[RRUR 5 [RRUR5 gU(a#[US 5(aUR5 ggg) zInitialize module weights. Args: module: PyTorch module to initialize. name: Module name. zero_init_last: Zero-initialize last layer weights. head.fcreg{Gz?)meanstdfan_outrelu)mode nonlinearityNrY) isinstancerHLinearConv2drUnormal_rWrVbiaskaiming_normal_ BatchNorm2d LayerNorm GroupNormones_hasattrrY)rrFrYs rPrrs&"))$$d):z&RTR[R[?\?\  CT: v{{# FBII & &  IFS ;; " GGNN6;; ' # FR^^R\\2<<H I I  fmm$ v{{# GF,<==>rRmodelrrc SSKnSnURU5n[URRR R SU"XRS355nURRR RU5 URR RU"XRS355 URRRU"XRS355 [[URSS5[R5(aURRR R SXRS3R S :XatURRR RU"XRS355 URRRRU"XRS 355 [!UR"R%55GHVunup[!U R&R%55GH%un upS n US US-S U S-SS3nU R(R RU"X^SU S355 U R*R RU"X^SU S355 U R,R RU"X^SU S355 U R.R RU"X^S355 U R0R RU"X^S355 U R2R RU"X^S355 U R.RRU"X^S355 U R0RRU"X^S355 U R2RRU"X^S355 U R4cGMX^SU S3nU R4RR RU"U55 GM( GMY g)NrctURS:XaUR/SQ5n[R"U5$)zPossibly convert HWIO to OIHW.r)r?rrr)ndim transposero from_numpy) conv_weightss rPt2p_load_weights..t2p7s1    !'11,?L --rRrz%root_block/standardized_conv2d/kernelzgroup_norm/gammazgroup_norm/betarzhead/conv2d/kernelzhead/conv2d/biasstandardized_conv2dblockz/unit02d/za/z/kernelzb/zc/za/group_norm/gammazb/group_norm/gammazc/group_norm/gammaza/group_norm/betazb/group_norm/betazc/group_norm/betaza/proj/)numpyloadr"rrrWr'copy_rrSrOrrrHrQrrrnamed_childrenrrErGryrDrFrxrC)rZrrnpraweights stem_conv_wisnamerjbnamerecname block_prefixws rPr r 3s. ggo&G" $$Q'Wx?d5e-f)giK JJOO  - JJC(2B(C DEF JJOO#g&@ABC'%**dD1299== JJMM & &q )Wx?Q5R-S-Y-YZ\-] ] ""3w9K/L'M#NO   Wx7G-H%I!JK&u||'B'B'DE>E!*5<<+F+F+H!I A~)E$XU1q5'q1uSkCL KK   $ $S>E7'1R)S%T U KK   $ $S>E7'1R)S%T U KK   $ $S>E7'1R)S%T U KK   $ $S>AS1T)U%V W KK   $ $S>AS1T)U%V W KK   $ $S>AS1T)U%V W KK   " "3w?P/Q'R#S T KK   " "3w?P/Q'R#S T KK   " "3w?P/Q'R#S T+nGE7'BC  %%,,223q6:"JFrRvariant pretrainedkwargsc :[SS9n[[X4SU0UD6$)zCreate a ResNetV2 model. Args: variant: Model variant name. pretrained: Load pretrained weights. **kwargs: Additional model arguments. Returns: ResNetV2 model instance. T)flatten_sequential feature_cfg)rrr&)rvrwrxr{s rP_create_resnetv2r|Ys2$/K '    rRc :[U4US[[SS9S.UD6$)zCreate a ResNetV2 model with BiT weights. Args: variant: Model variant name. pretrained: Load pretrained weights. **kwargs: Additional model arguments. Returns: ResNetV2 model instance. rg:0yE>)eps)rwrr2)r|rr)rvrwrxs rP_create_resnetv2_bitrls3  9$/     rRurlc 2USSSSS[[SSSS . UE$) Nr)r?r)rrg?bilinearrrHz apache-2.0) rr input_size pool_sizecrop_pct interpolationrIrJ first_conv classifierlicenser )rrxs rP_cfgrs3=vJ'0F!    rRz%resnetv2_50x1_bit.goog_distilled_in1kztimm/bicubic) hf_hub_idr custom_loadz-resnetv2_152x2_bit.goog_teacher_in21k_ft_in1kz1resnetv2_152x2_bit.goog_teacher_in21k_ft_in1k_384)r?r) rrd)rrrrrrz$resnetv2_50x1_bit.goog_in21k_ft_in1k)r?r)r)rrrrrz$resnetv2_50x3_bit.goog_in21k_ft_in1kz%resnetv2_101x1_bit.goog_in21k_ft_in1kz%resnetv2_101x3_bit.goog_in21k_ft_in1kz%resnetv2_152x2_bit.goog_in21k_ft_in1kz%resnetv2_152x4_bit.goog_in21k_ft_in1k)r?r)rzresnetv2_50x1_bit.goog_in21kiSU)rrrzresnetv2_50x3_bit.goog_in21kzresnetv2_101x1_bit.goog_in21kzresnetv2_101x3_bit.goog_in21kzresnetv2_152x2_bit.goog_in21kzresnetv2_152x4_bit.goog_in21kzresnetv2_18.ra4_e3600_r224_in1kg?)r? r)rrrtest_input_size test_crop_pctz resnetv2_18d.ra4_e3600_r224_in1kz stem.conv1)rrrrrr)rrrrrrrgffffff?)rr)r)rrrrrr) zresnetv2_34.ra4_e3600_r224_in1kz resnetv2_34d.ra4_e3600_r224_in1kz resnetv2_34d.ra4_e3600_r384_in1kzresnetv2_50.a1h_in1kzresnetv2_50d.untrainedzresnetv2_50t.untrainedzresnetv2_101.a1h_in1kzresnetv2_101d.untrainedzresnetv2_152.untrainedzresnetv2_152d.untrainedzresnetv2_50d_gn.ah_in1kzresnetv2_50d_evos.ah_in1kzresnetv2_50d_frn.untrainedc &[SU/SQSS.UD6$)zResNetV2-50x1-BiT model.r?rr?rrwrr)resnetv2_50x1_bitrrwrxs rPrr-  c(2&9TW ^J  _j _DD^W]D^ __rRc d[/SQSSS[[SSS9n[SSU0[U40UD6D6$) z'ResNetV2-18d model (deep stem variant).rrTrdrrrrr,r2r3rrrw) resnetv2_18drrs rPrr$sC&9TW ^vX\J  `z `T*E_X^E_ ``rRc \[SSSS[[S9n[SSU0[U40UD6D6$)zResNetV2-34 model.rrTrdrrw) resnetv2_34rrs rPrr.s>&9TW ^J  _j _DD^W]D^ __rRc `[SSSS[[SSS9n[SSU0[U40UD6D6$) z'ResNetV2-34d model (deep stem variant).rrTrdrrrw) resnetv2_34drrs rPrr8sC&9TW ^vX\J  `z `T*E_X^E_ ``rRc Z[/SQ[[S9n[SSU0[U40UD6D6$)zResNetV2-50 model.rrr2r3rw) resnetv2_50rrs rPrrBs1\mP^_J  _j _DD^W]D^ __rRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z'ResNetV2-50d model (deep stem variant).rrTrr2r3rrrw) resnetv2_50drrs rPrrIs: .4)J  `z `T*E_X^E_ ``rRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z)ResNetV2-50t model (tiered stem variant).rrTrrw) resnetv2_50trrs rPrrRs: .T+J  `z `T*E_X^E_ ``rRc Z[/SQ[[S9n[SSU0[U40UD6D6$)zResNetV2-101 model.rrrw) resnetv2_101rrs rPrr[1]}Q_`J  `z `T*E_X^E_ ``rRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z(ResNetV2-101d model (deep stem variant).rrTrrw) resnetv2_101drrs rPrrb:>4)J  a ad:F`Y_F` aarRc Z[/SQ[[S9n[SSU0[U40UD6D6$)zResNetV2-152 model.rrrw) resnetv2_152rrs rPrrkrrRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z(ResNetV2-152d model (deep stem variant).rrTrrw) resnetv2_152drrs rPrrrrrRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z,ResNetV2-50d model with Group Normalization.rrTrrw)resnetv2_50d_gn)rrrr|rs rPrr}s: ,4)J  c* cZHb[aHb ccrRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z ResNetV2-50d model with EvoNorm.rrTrrw)resnetv2_50d_evos)rrrr|rs rPrrs: +4)J  eJ e$zJd]cJd eerRc ^[/SQ[[SSS9n[SSU0[U40UD6D6$)z6ResNetV2-50d model with Filter Response Normalization.rrTrrw)resnetv2_50d_frn)rrrr|rs rPrrs; BY4)J  d: djIc\bIc ddrR)resnetv2_50x1_bitmresnetv2_50x3_bitmresnetv2_101x1_bitmresnetv2_101x3_bitmresnetv2_152x2_bitmresnetv2_152x4_bitmresnetv2_50x1_bitm_in21kresnetv2_50x3_bitm_in21kresnetv2_101x1_bitm_in21kresnetv2_101x3_bitm_in21kresnetv2_152x2_bitm_in21kresnetv2_152x4_bitm_in21kresnetv2_50x1_bit_distilledresnetv2_152x2_bit_teacherresnetv2_152x2_bit_teacher_384)rTrBrC)r)Yrl collectionsr functoolsrtypingrrrrr r r rotorch.nnrH timm.datar r timm.layersrrrrrrrrrrrrrr_builderr _featuresr _manipulater r!r" _registryr#r$r%__all__rEr(rurrrrrrrrmrrr&rno_gradr r|rr default_cfgsrrrrrrrrrrrrrrrrrrrrrhrRrPrs >$DDD ESSSS*+FFYY ,Y"))YxbryybJPPf'RYY'B$2BII$2N<"))<~ =C =D =(&|C---- -  -  -]]-`QryyQh ")) 3 T UY ,";";S";#";";Jctsx&#43S[( c # $sCx. %a&+TT.3a& 4TT63 a&8 HsR[im:oa&+D HsPT-Va& +D HsPT-V!a&&,T HsPT.V'a&,,T HsPT.V-a&2,T HsPT.V3a&8,T HsPT.V9a&B#Dt%-Ca&H#Dt%-Ia&N$Tt&-Oa&T$Tt&-Ua&Z$Tt&-[a&`$Tt&-aa&h&t#}\_(aia&n'#}\_)!oa&v(,#}\_(a)-#}\_)!)-(TaL):!$ ]`b#L:"L:!$ ]`b $L :"!#L : $L}C I"&L}C"I#'L#:a&a Hc$c#c(cc c$c#c(cc e4e3e8ee e4e3e8ee e4e3e8ee e4e3e8ee `D`C`H``aTaSaXaa`D`C`H``aTaSaXaa`D`C`H`` aTaSaXaaaTaSaXaaaTaSaXaa bdbcbhbbaTaSaXaa bdbcbhbbdddddf$f#f(ffeeeeeH@@BBBB > >!@!@!@!@#J"Q&Y'rR