hSSKrSSKJr SSKJrJrJr SSKrSSKJ s J r SSKJ r J r SSK JrJr SSKJr SSKJrJr SS KJr S S KJrJrJr S S KJr S S KJrJr /SQr S\RS\R4Sjr!\RDRGS5 S\RS\RS\$\%S\R4Sjr&\RDRGS5 "SS\ RN5r("SS\ RN5r)SSS\ S\ S\ S \ S\$\%S!\%S"\$\%S#\*S$\*S%\\ S&\\ S'\\RS(\+S\ 4S)jjr,\RDRGS*5 "S+S,\ RN5r-"S-S.\-5r."S/S0\ RN5r/"S1S2\/5r0"S3S4\ RN5r1S5\$\%S6\%S7\$\%S!\$\%S\$\%S8\*S9\\S:\+S;\S\14S<jr2S=\0r3"S>S?\5r4"S@SA\5r5"SBSC\5r6"SDSE\5r7"SFSG\5r8"SHSI\5r9\"5\"SJ\4Rt4SK9SSSL.S9\\4S:\+S;\S\14SMjj55r;\"5\"SJ\5Rt4SK9SSSL.S9\\5S:\+S;\S\14SNjj55r<\"5\"SJ\6Rt4SK9SSSL.S9\\6S:\+S;\S\14SOjj55r=\"5\"SJ\7Rt4SK9SSSL.S9\\7S:\+S;\S\14SPjj55r>\"5\"SJ\8Rt4SK9SSSL.S9\\8S:\+S;\S\14SQjj55r?\"5\"SJ\9Rt4SK9SSSL.S9\\9S:\+S;\S\14SRjj55r@g)TN)partial)AnyCallableOptional)nnTensor)MLPPermute)StochasticDepth)ImageClassificationInterpolationMode)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) SwinTransformerSwin_T_WeightsSwin_S_WeightsSwin_B_WeightsSwin_V2_T_WeightsSwin_V2_S_WeightsSwin_V2_B_Weightsswin_tswin_sswin_b swin_v2_t swin_v2_s swin_v2_bxreturnc *URSSupn[R"USSSUS-SUS-45nUSSSS2SSS2SS24nUSSSS2SSS2SS24nUSSSS2SSS2SS24nUSSSS2SSS2SS24n[R"XEXg/S5nU$)Nrr .r)shapeFpadtorchcat)r$HW_x0x1x2x3s [/var/www/auris/envauris/lib/python3.13/site-packages/torchvision/models/swin_transformer.py_patch_merging_padr6#sggbclGA! a!Q1q5!QU+,A 31addA  B 31addA  B 31addA  B 31addA  B 22"B'A Hr6relative_position_bias_tablerelative_position_index window_sizecUSUS-nXnURX3S5nURSSS5R5RS5nU$)Nrrr(r )viewpermute contiguous unsqueeze)r8r9r:Nrelative_position_biass r5_get_relative_position_biasrB1sb AQ'A9R388rB3;;Aq!DOOQ[[\]^ !!r7rBc~^\rSrSrSr\R 4S\S\S\R44U4Sjjjr S\ 4Sjr S r U=r$) PatchMerging>zPatch Merging Layer. Args: dim (int): Number of input channels. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. dim norm_layer.c>[TU]5 [U5 Xl[R "SU-SU-SS9UlU"SU-5UlgNr Fbiassuper__init__rrFrLinear reductionnormselfrFrG __class__s r5rOPatchMerging.__init__EG D!1s7AG%@q3w' r7r$c`[U5nURU5nURU5nU$z Args: x (Tensor): input tensor with expected layout of [..., H, W, C] Returns: Tensor with layout of [..., H/2, W/2, 2*C] )r6rRrQrTr$s r5forwardPatchMerging.forwardLs. q ! IIaL NN1 r7rFrRrQ__name__ __module__ __qualname____firstlineno____doc__r LayerNormintrModulerOrr[__static_attributes__ __classcell__rUs@r5rDrD>I IK (C(Xc299n-E((   r7rDc~^\rSrSrSr\R 4S\S\S\R44U4Sjjjr S\ 4Sjr S r U=r$) PatchMergingV2YzPatch Merging Layer for Swin Transformer V2. Args: dim (int): Number of input channels. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. rFrG.c>[TU]5 [U5 Xl[R "SU-SU-SS9UlU"SU-5UlgrIrMrSs r5rOPatchMergingV2.__init__`rWr7r$c`[U5nURU5nURU5nU$rY)r6rQrRrZs r5r[PatchMergingV2.forwardgs. q ! NN1  IIaLr7r]r^ris@r5rlrlYrjr7rlTinput qkv_weight proj_weightrA num_heads shift_sizeattention_dropoutdropoutqkv_bias proj_bias logit_scaletrainingc : URupnnUSXS-- US-nUSXS-- US-n[R"USSSUSU45nURunnnnUR5nUSU:aSUS'USU:aSUS'[ U5S:a [ R "UUS*US*4SS9nUUS-UUS--nURU UUS-USUUS-USU5nURSSSSSS5RU U-USUS-U5nU b<U b9U R5n U R5S-nU USU-R5 [R"UX5nURURS5URS5SUUU-5RSSSSS5nUSUSUSnnnU bx[R"US S 9[R"US S 9R!S S 5-n[ R""U [$R&"S 5S9R)5n UU -nO,UUU-S--nUR+UR!S S 55nUU-n[ U5S:GaUR-UU45nSUS*4US*US*4US*S 44nSUS*4US*US*4US*S 44n Sn!UH(n"U Hn#U!UU"SU"S2U#SU#S24'U!S- n!M! M* URUUS-USUUS-US5nURSSSS5RUUSUS-5nUR/S5UR/S5- nUR1US:g[3S55R1US:H[3S55nURURS5U-UUURS5URS55nUUR/S5R/S5-nURS UURS5URS55n[R4"US S 9n[R6"UX|S9nUR+U5R!SS5RURS5URS5U5n[R"UX*5n[R6"UXS9nURU UUS-UUS-USUSU5nURSSSSSS5RU UUU5n[ U5S:a[ R "UUSUS4SS9nUS S 2S U2S U2S S 24R95nU$)a Window based multi-head self attention (W-MSA) module with relative position bias. It supports both of shifted and non-shifted window. Args: input (Tensor[N, H, W, C]): The input tensor or 4-dimensions. qkv_weight (Tensor[in_dim, out_dim]): The weight tensor of query, key, value. proj_weight (Tensor[out_dim, out_dim]): The weight tensor of projection. relative_position_bias (Tensor): The learned relative position bias added to attention. window_size (List[int]): Window size. num_heads (int): Number of attention heads. shift_size (List[int]): Shift size for shifted window attention. attention_dropout (float): Dropout ratio of attention weight. Default: 0.0. dropout (float): Dropout ratio of output. Default: 0.0. qkv_bias (Tensor[out_dim], optional): The bias tensor of query, key, value. Default: None. proj_bias (Tensor[out_dim], optional): The bias tensor of projection. Default: None. logit_scale (Tensor[out_dim], optional): Logit scale of cosine attention for Swin Transformer V2. Default: None. training (bool, optional): Training flag used by the dropout parameters. Default: True. Returns: Tensor[N, H, W, C]: The output tensor after shifted window attention. rr)rr )shiftsdimsr rJNr()rFgY@)maxggY)pr|)r)r*r+copysumr,rollr<r=reshapeclonenumelzero_linearsize normalize transposeclampmathlogexpmatmul new_zerosr? masked_fillfloatsoftmaxrxr>)$rrrsrtrAr:rurvrwrxryrzr{r|Br.r/Cpad_rpad_br$r0pad_Hpad_W num_windowslengthqkvqkvattn attn_maskh_slicesw_slicescounthws$ r5shifted_window_attentionrtsFJA!Q ^aa.0 0KN BE ^aa.0 0KN BE eaAua/0AAueQ"J1~ 1 1~ 1  : JJq:a=.:a=.!A OKN*u A/FGK q%;q>);q>5KPQN;RT_`aTbdefA !Q1a#++A O[^kZ[n=\^_`A8#7>>#!Q&!f*%++- ((1j +C ++affQiAy!y. I Q QRSUVXY[\^_ `C!fc!fc!f!qA{{1"% A2(>(H(HR(PPkk+488E?CGGI k! iD( (xx B+, ( (D :KK/ Q(KN?Z]N*KzZ[}n^bMcdQ(KN?Z]N*KzZ[}n^bMcdA6; !A$1+qtad{23 NN5KN#:KNEU`abUcLcepqrest %%aAq199+{ST~XcdeXfGfg ''*Y-@-@-CC )))q.%-HTTU^bcUcejkneop yyk1; 166RS9VWV\V\]^V_`i))!,66q99yyYq 166!9= 99Tr "D 99T. BD A  A&..qvvay!&&)QGA K+A !w2A q%;q>)5KN+BKPQNT_`aTbdefA !Q1a#++AueQ?A : JJq*Q-A!?f M !RaR!Q,""$A Hr7rc^\rSrSrSrSS\S\\S\\S\S\S\S \S \4U4S jjjr S r S r S\ R4SjrS\ S\ 4SjrSrU=r$)ShiftedWindowAttentionz' See :func:`shifted_window_attention`. rFr:rvruryrzrwrxc ^>[T U]5 [U5S:wd[U5S:wa [S5eX lX0lX@lXplXl[R"XS-US9Ul [R"XUS9Ul UR5 UR5 g)Nr z.window_size and shift_size must be of length 2rrK)rNrOlen ValueErrorr:rvrurwrxrrPrproj#define_relative_position_bias_tabledefine_relative_position_index) rTrFr:rvruryrzrwrxrUs r5rOShiftedWindowAttention.__init__s  { q C Oq$8MN N&$"!2 99S'9IIcY7  002 ++-r7c[R"[R"SURS-S- SURS-S- -UR 55Ul[RRUR SS9 g)Nr rr{Gz?std) r Parameterr,zerosr:rur8init trunc_normal_rTs r5r:ShiftedWindowAttention.define_relative_position_bias_table sw,.LL KKT--a0014T=M=Ma=P9PST9TUW[WeWe f- ) d??TJr7c[R"URS5n[R"URS5n[R"[R"XSS95n[R "US5nUSS2SS2S4USS2SSS24- nUR SSS5R5nUSS2SS2S4==URSS- - ss'USS2SS2S4==URSS- - ss'USS2SS2S4==SURS-S- -ss'URS5R 5nURSU5 g)Nrrijindexingr r(r9) r,aranger:stackmeshgridflattenr=r>rregister_buffer)rTcoords_hcoords_wcoordscoords_flattenrelative_coordsr9s r5r5ShiftedWindowAttention.define_relative_position_indexs=<< 0 0 34<< 0 0 34U^^HNOvq1(At4~aqj7QQ)11!Q:EEG1a D$4$4Q$7!$;; 1a D$4$4Q$7!$;; 1a A(8(8(;$;a$?? "1"5"5b"9"A"A"C 68OPr7r%cX[URURUR5$N)rBr8r9r:rs r5get_relative_position_bias1ShiftedWindowAttention.get_relative_position_biass(*  - -t/K/KTM]M]  r7r$cbUR5n[UURRURRUUR UR URURURURRURRURS9 $){ Args: x (Tensor): Tensor with layout of [B, H, W, C] Returns: Tensor with same layout as input, i.e. [B, H, W, C] )rvrwrxryrzr|) rrrweightrr:rurvrwrxrLr|rTr$rAs r5r[ShiftedWindowAttention.forward#s"&!@!@!B' HHOO II   "    NN"44LLXX]]iinn]]  r7)rwrxrurrr8rvr:TTrr)r_r`rarbrcrelistboolrrOrrr,rrr[rgrhris@r5rrs#&. .#Y.I .  .  ..!...4K Q ELL   F  r7rc^\rSrSrSrSS\S\\S\\S\S\S\S \S \4U4S jjjr S r S \ R4Sjr S\ 4SjrSrU=r$)ShiftedWindowAttentionV2i;z* See :func:`shifted_window_attention_v2`. rFr:rvruryrzrwrxc 4>[T U]UUUUUUUUS9 [R"[R "S[R "USS45-55Ul[R"[R"SSSS9[R"SS9[R"SUS S95Ul U(a\URRR5S -n URRU SU -RR!5 gg) N)ryrzrwrx rr iTrK)inplaceFr)rNrOrrr,ronesr{ SequentialrPReLUcpb_mlprrLrdatar) rTrFr:rvruryrzrwrxrrUs r5rO!ShiftedWindowAttentionV2.__init__@s     /  << "uzz9aQRBS7T2T(UV}} IIa4 ("''$*?3PY`eAf   XX]]((*a/F HHMM&1v: . 3 3 9 9 ; r7c[R"URSS- *URS[RS9n[R"URSS- *URS[RS9n[R"[R "X/SS95nUR SSS5R5RS5nUSS2SS2SS2S4==URSS- -ss'USS2SS2SS2S4==URSS- -ss'US-n[R"U5[R"[R"U5S-5-S - nURS U5 g) Nrr)dtyperrr g?g@relative_coords_table) r,rr:float32rrr=r>r?signlog2absr)rTrelative_coords_hrelative_coords_wrs r5r+B)CTEUEUVWEX`e`m`mn!LL4+;+;A+>+B)CTEUEUVWEX`e`m`mn % ENNM &ELL&   r7rc^\rSrSrSrSSSS\R \4S\S\S\ \S\ \S \ S \ S \ S \ S \ S\R4S\ S\R44U4Sjjjr S\4SjrSrU=r$)SwinTransformerBlockia Swin Transformer Block. Args: dim (int): Number of input channels. num_heads (int): Number of attention heads. window_size (List[int]): Window size. shift_size (List[int]): Shift size for shifted window attention. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.0. dropout (float): Dropout rate. Default: 0.0. attention_dropout (float): Attention dropout rate. Default: 0.0. stochastic_depth_prob: (float): Stochastic depth rate. Default: 0.0. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. attn_layer (nn.Module): Attention layer. Default: ShiftedWindowAttention @rrFrur:rv mlp_ratiorxrwstochastic_depth_probrG. attn_layerc L>[T U]5 [U5 U "U5UlU "UUUUUUS9Ul[ US5UlU "U5Ul[U[X-5U/[RSUS9Ul URR5Hn [U [R5(dM$[R R#U R$5 U R&cM\[R R)U R&SS9 M g)N)rwrxrow)activation_layerrrxgư>r)rNrOrnorm1rr stochastic_depthnorm2r rerGELUmlpmodules isinstancerPrxavier_uniform_rrLnormal_) rTrFrur:rvrrxrwrrGrmrUs r5rOSwinTransformerBlock.__init__s D!_     /   !00Eu M_ sS137"''[_ipq!!#A!RYY''''166%GGOOAFFO5 $r7r$cXRURURU555-nXRURUR U555-nU$r)rrrrrrZs r5r[SwinTransformerBlock.forwardsO %%dii 1 &>? ? %%dhhtzz!}&=> >r7)rrrrr)r_r`rarbrcrrdrrerrrrfrOrr[rgrhris@r5rrs *#&'*/1||/E!6 !6!6#Y !6 I !6  !6!6!!6 %!6S"))^,!6S"))^,!6!6Fr7rc^\rSrSrSrSSSS\R \4S\S\S\ \S\ \S \ S \ S \ S \ S \ S\R4S\ S\R44U4Sjjjr S\4SjrSrU=r$)SwinTransformerBlockV2ia Swin Transformer V2 Block. Args: dim (int): Number of input channels. num_heads (int): Number of attention heads. window_size (List[int]): Window size. shift_size (List[int]): Shift size for shifted window attention. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.0. dropout (float): Dropout rate. Default: 0.0. attention_dropout (float): Attention dropout rate. Default: 0.0. stochastic_depth_prob: (float): Stochastic depth rate. Default: 0.0. norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm. attn_layer (nn.Module): Attention layer. Default: ShiftedWindowAttentionV2. rrrFrur:rvrrxrwrrG.rc 2>[T U]UUUUUUUUU U S9 g)N)rrxrwrrGr)rNrO) rTrFrur:rvrrxrwrrGrrUs r5rOSwinTransformerBlockV2.__init__s5     /"7!!  r7r$cXRURURU555-nXRURUR U555-nU$r)rrrrrrZs r5r[SwinTransformerBlockV2.forwardsQ %%djj1&>? ? %%djj!&=> >r7)r_r`rarbrcrrdrrerrrrfrOrr[rgrhris@r5rrs *#&'*/1||/G   #Y  I     !  % S"))^, S"))^,  4r7rc^\rSrSrSrSSSSSSS\4S\\S \S \\S \\S \\S \S\S\S\S\S\ \ S\ R4S\ \ S\ R4S\ S\ R44U4Sjjjr SrSrU=r$)ria Implements Swin Transformer from the `"Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" `_ paper. Args: patch_size (List[int]): Patch size. embed_dim (int): Patch embedding dimension. depths (List(int)): Depth of each Swin Transformer layer. num_heads (List(int)): Number of attention heads in different layers. window_size (List[int]): Window size. mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.0. dropout (float): Dropout rate. Default: 0.0. attention_dropout (float): Attention dropout rate. Default: 0.0. stochastic_depth_prob (float): Stochastic depth rate. Default: 0.1. num_classes (int): Number of classes for classification head. Default: 1000. block (nn.Module, optional): SwinTransformer Block. Default: None. norm_layer (nn.Module, optional): Normalization layer. Default: None. downsample_layer (nn.Module): Downsample layer (patch merging). Default: PatchMerging. rrg?iN patch_size embed_dimdepthsrur:rrxrwr num_classesrG.blockdownsample_layerc>[TU]5 [U5 XlU c[n U c[ [ RSS9n /nUR[ R"[ R"SX!SUS4USUS4S9[/SQ5U "U555 [U5nSn[[U55Hn/nUSU--n[UU5H[nU [U5-US- - nURU "UUUUUVs/sHnUS-S:XaSOUS-PM snUUUUU S 9 5 US- nM] UR[ R"U65 U[U5S- :dMURU "UU 55 M [ R"U6UlUS[U5S- --nU "U5Ul[/S Q5Ul[ R&"S5Ul[ R*"S5Ul[ R."UU 5UlUR35Hn[5U[ R.5(dM$[ R6R9UR:S S 9 UR<cM[[ R6R?UR<5 M gs snf) Ngh㈵>)epsrrr) kernel_sizestride)rr rrr )r:rvrrxrwrrG)rrrr rr) rNrOrrrrrrdappendrConv2dr rrangerrfeaturesrRr=AdaptiveAvgPool2davgpoolFlattenrrPheadrrrrrrLzeros_)rTrrrrur:rrxrwrrrGrrlayerstotal_stage_blocksstage_block_idi_stagestagerFi_layersd_probr num_featuresr rUs r5rOSwinTransformer.__init__sw D!& =(E   48J"$ MM y!}jm.LV`abVceopqerUs %9%   ![S[)G%'Eaj(C 1/%2GGK]`aKab !'*$/OZ#[{!1)9AqAv$E{#["+ '*;.5#-  !#!2" MM"--/ 0#f+/* .sJ?@/*0 v.  1Vq#99 |, |, ++A. zz!} IIlK8 A!RYY''%%ahhD%966%GGNN166*  -$\s KcURU5nURU5nURU5nURU5nUR U5nUR U5nU$r)r$rRr=r&rr(rZs r5r[SwinTransformer.forward_sV MM!  IIaL LLO LLO LLO IIaLr7)r&r$rr(rRrr=)r_r`rarbrcrDrrerrrrrfrOr[rgrhris@r5rrs4#&'*9=485AM+IM+M+S M+ 9 M+ #Y M+M+M+!M+ %M+M+Xc299n56M+bii01M+#3 >2M+M+^r7rrrrrweightsprogresskwargsc Ub#[US[URS55 [SUUUUUUS.UD6n UbU R UR USS95 U $)Nr categories)rrrrur:rT)r6 check_hashr)rrmetarload_state_dictget_state_dict) rrrrur:rr5r6r7models r5_swin_transformerr?is{fmSl9S5TU  3    E g44hSW4XY Lr7r9c l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)riz7https://download.pytorch.org/models/swin_t-704ceda3.pth crop_size resize_size interpolationibrArAUhttps://github.com/pytorch/vision/tree/main/references/classification#swintransformer ImageNet-1KguV^T@glW@zacc@1zacc@5gX9@g\( [@YThese weights reproduce closely the results of the paper using a similar training recipe. num_paramsmin_sizerecipe_metrics_ops _file_size_docsurl transformsr;rN r_r`rarbrrr rBICUBIC _COMMON_META IMAGENET1K_V1DEFAULTrgrr7r5rrsg E 3CO`OhOh   ""m##   t M*Gr7rc l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)riz7https://download.pytorch.org/models/swin_s-5e29d889.pthrArCirrGrHrIgCT@gףp= X@rJgZd{!@gx&g@rKrLrTrNrWrr7r5rrsg E 3CO`OhOh   ""m##  !t M*Gr7rc l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)riz7https://download.pytorch.org/models/swin_b-68c6b09e.pthrArCi<;rGrHrIgh|?T@g)\(X@rJg&1.@gt@rKrLrTrNrWrr7r5rrsg E 3CO`OhOh   ""m##  !t M*Gr7rc l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)riz:https://download.pytorch.org/models/swin_v2_t-b137f0e2.pthrCiRraraXhttps://github.com/pytorch/vision/tree/main/references/classification#swintransformer-v2rIgS㥛T@g rX@rJg(\@gMb([@rKrLrTrNrWrr7r5rrsg H 3CO`OhOh   ""p##  !t M*Gr7rc l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)riz:https://download.pytorch.org/models/swin_v2_s-637d8ceb.pthrarbrCircrdrIg!rhT@gNbX94X@rJgd;O'@gg@rKrLrTrNrWrr7r5rrg H 3CO`OhOh   ""p##  !t M*Gr7rc l\rSrSr\"S\"\SS\RS90\ ESSSS S S S .0S SSS.ES9r \ r Sr g)ri z:https://download.pytorch.org/models/swin_v2_b-781e5279.pthrairCi=rcrdrIgI +U@gK7X@rJg33333S4@gˡEu@rKrLrTrNrWrr7r5rr rfr7r pretrained)r5)r5r6c d[RU5n[SSS/S/SQ/SQSS/SUUS.UD6$) a Constructs a swin_tiny architecture from `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows `_. Args: weights (:class:`~torchvision.models.Swin_T_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_T_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_T_Weights :members: rJ`r r r rrl 皙?rrrrur:rr5r6r)rverifyr?r5r6r7s r5rr"sP.##G,G  q6 F!    r7c d[RU5n[SSS/S/SQ/SQSS/SUUS.UD6$) a Constructs a swin_small architecture from `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows `_. Args: weights (:class:`~torchvision.models.Swin_S_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_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.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_S_Weights :members: rJrjr r r rmrp333333?rrr)rrsr?rts r5rrHsP.##G,G  q6 F!    r7c d[RU5n[SSS/S/SQ/SQSS/SUUS.UD6$) a Constructs a swin_base architecture from `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows `_. Args: weights (:class:`~torchvision.models.Swin_B_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_B_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_B_Weights :members: rJrvrJrr rp?rrr)rrsr?rts r5r r nsP.##G,G  q6 F!    r7c x[RU5n[SSS/S/SQ/SQSS/SUU[[S. UD6$) a| Constructs a swin_v2_tiny architecture from `Swin Transformer V2: Scaling Up Capacity and Resolution `_. Args: weights (:class:`~torchvision.models.Swin_V2_T_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_V2_T_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.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_V2_T_Weights :members: rJrjrkrmrrq rrrrur:rr5r6rrr)rrsr?rrlrts r5r!r!sV. &&w/G  q6 F!$'    r7c x[RU5n[SSS/S/SQ/SQSS/SUU[[S. UD6$) a} Constructs a swin_v2_small architecture from `Swin Transformer V2: Scaling Up Capacity and Resolution `_. Args: weights (:class:`~torchvision.models.Swin_V2_S_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_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.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_V2_S_Weights :members: rJrjrvrmrrxrr)rrsr?rrlrts r5r"r"sV. &&w/G  q6 F!$'    r7c x[RU5n[SSS/S/SQ/SQSS/SUU[[S. UD6$) a| Constructs a swin_v2_base architecture from `Swin Transformer V2: Scaling Up Capacity and Resolution `_. Args: weights (:class:`~torchvision.models.Swin_V2_B_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.Swin_V2_B_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.swin_transformer.SwinTransformer`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.Swin_V2_B_Weights :members: rJrzrvr{rr}rr)rrsr?rrlrts r5r#r#sV. &&w/G  q6 F!$'    r7)rrNNNT)Ar functoolsrtypingrrrr,torch.nn.functionalr functionalr*rops.miscr r ops.stochastic_depthr transforms._presetsr rutilsr_apirrr_metar_utilsrr__all__r6fxwraprrerBrfrDrlrrrrrrrrr?rYrrrrrrrZrrr r!r"r#rr7r5rs ** #2H'66'B " %,, 5<<  "#""',,"IN"dhildm" \\" +,2996RYYF #!%"&*.p p p p # p c p  p S p p p vp p %,,'p p  p f ()N RYYN bS 5S l62996r/1/djbiijZS  ICy  c  ! k ">& [2[2[2 2 2 2,0L0L!MN26!x/!$!Y\!ap!O!H,0L0L!MN26!x/!$!Y\!ap!O!H,0L0L!MN26!x/!$!Y\!ap!O!H,0A0O0O!PQ8