+ jBRt^RIt^RIt^RIHtHt^RIHt^RIH t ^RI H t H t ^RI HtHtHtHt^RIHt^RIHtHt^R IHt^R IHt]P6!]4t!R R ]P<4t!R R]P<4t !RR]P<4t!!RR]P<4t"!RR]P<4t#!RR]P<4t$!RR]P<4t%]!RR]44t&]!RR]&44t']!RR7!RR ]&44t(]!R!R7!R"R#] ]&44t).R$Ot*R#)%zPyTorch ResNet model.N)Tensornn)initialization)ACT2FN) BackboneMixinfilter_output_hidden_states)BackboneOutputBaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)auto_docstringlogging)can_return_tuple ResNetConfigcTaa]tRt^'toRV3RlV3RllltV3RlRltRtVtV;t#)ResNetConvLayerc<V^8dQhRS[RS[RS[S[S[S[3,,RS[RS[RS[S[S[S[3,,RS[RS[/#)  in_channels out_channels kernel_sizestridebiasdilationgroups activation)inttupleboolstr)format __classdict__s"k/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/resnet/modeling_resnet.py __annotate__ResNetConvLayer.__annotate__(sZZZZ5c?* Z  Z  Zc3h'ZZZc <\S V`4\P!VVVVV^,VVVR7Vn\P !V4VnVe\V,Vn R#\P!4Vn R#)r)rrrrpaddingrrrN) super__init__rConv2d convolution BatchNorm2d normalizationrIdentityr) selfrrrrrrrr __class__s &&&&&&&&&r$r+ResNetConvLayer.__init__(sn 99#%#1$   ^^L90:0F&,BKKMr'cN<V^8dQhRS[PRS[P/#)r hidden_statesreturn)torchr)r"r#s"r$r%r&As#U\\ellr'clVPV4pVPV4pVPV4pV#N)r-r/r)r1r5s&&r$forwardResNetConvLayer.forwardAs6((7 **=9  6 r')rr-r/)Fr=r=relu __name__ __module__ __qualname____firstlineno__r+r:__static_attributes____classdictcell__ __classcell__r2r#s@@r$rr's ZZ2r'rcTaa]tRt^HtoRtV3RlV3RlltV3RlRltRtVtV;t #)ResNetEmbeddingszG ResNet Embeddings (stem) composed of a single aggressive convolution. c <V^8dQhRS[/#rconfigr)r"r#s"r$r%ResNetEmbeddings.__annotate__Ms00|0r'c<\SV`4\VPVP^^VP R7Vn\P!^^^R7Vn VPVnR#))rrr)rrr)N) r*r+r num_channelsembedding_size hidden_actembedderr MaxPool2dpoolerr1rLr2s&&r$r+ResNetEmbeddings.__init__Ms\ '   !6!6Aa\b\m\m  llqAF "//r'c&<V^8dQhRS[RS[/#)r pixel_valuesr6r)r"r#s"r$r%rMUsFvr'cVP^,pW P8wd \R4hVPV4pVP V4pV#)r=zeMake sure that the channel dimension of the pixel values match with the one set in the configuration.)shaperP ValueErrorrSrU)r1rYrP embeddings&& r$r:ResNetEmbeddings.forwardUsR#))!, ,, ,w MM,/ KK * r')rSrPrU r@rArBrC__doc__r+r:rDrErFrGs@@r$rIrIHs#00r'rIcXaa]tRt^`toRtRV3RlV3RllltV3RlRltRtVtV;t #)ResNetShortCutz ResNet shortcut, used to project the residual features to the correct size. If needed, it is also used to downsample the input using `stride=2`. c,<V^8dQhRS[RS[RS[/#)rrrr)r)r"r#s"r$r%ResNetShortCut.__annotate__fs"::C:s:C:r'c<\SV`4\P!W^VRR7Vn\P !V4VnR#)r=F)rrrN)r*r+rr,r-r.r/)r1rrrr2s&&&&r$r+ResNetShortCut.__init__fs8 99[AV\chi^^L9r'c&<V^8dQhRS[RS[/#rinputr6rZ)r"r#s"r$r%reksVr'cJVPV4pVPV4pV#r9r-r/)r1rj hidden_states&& r$r:ResNetShortCut.forwardks(''. )),7 r'rl)rr`rGs@@r$rcrc`s# :: r'rccLaa]tRt^qtoRtRV3RlV3RllltRtRtVtV;t #)ResNetBasicLayerzG A classic ResNet's residual layer composed by two `3x3` convolutions. c2<V^8dQhRS[RS[RS[RS[/#)rrrrr)rr!)r"r#s"r$r%ResNetBasicLayer.__annotate__vs* - -C -s -C -Y\ -r'c ,<\SV`4W8g;'gV^8gpV'd\WVR7M\P!4Vn\P !\WVR7\W"RR74Vn\V,Vn R#)r=rNr r*r+rcrr0shortcut Sequentialrlayerrr)r1rrrrshould_apply_shortcutr2s&&&&& r$r+ResNetBasicLayer.__init__vsv  + ; J Jv{H]N;V Dcecncncp ]] Kf E L4 H  !,r'cTpVPV4pVPV4pW, pVPV4pV#r9ryrwrr1rmresiduals&& r$r:ResNetBasicLayer.forward?zz,/ ==* |4 r'rryrw)r=r>r`rGs@@r$rprpqs - -r'rpcLaa]tRt^toRtRV3RlV3RllltRtRtVtV;t #)ResNetBottleNeckLayera A classic ResNet's bottleneck layer composed by three `3x3` convolutions. The first `1x1` convolution reduces the input by a factor of `reduction` in order to make the second `3x3` convolution faster. The last `1x1` convolution remaps the reduced features to `out_channels`. If `downsample_in_bottleneck` is true, downsample will be in the first layer instead of the second layer. c ><V^8dQhRS[RS[RS[RS[RS[RS[/#)rrrrr reductiondownsample_in_bottleneck)rr!r )r"r#s"r$r%"ResNetBottleNeckLayer.__annotate__sG---- -  -  -#'-r'c <\S V`4W8g;'gV^8gpW%,pV'd\WVR7M\P!4Vn\P !\Y^V'dTM^R7\YV'gTM^R7\W^RR74Vn\V,Vn R#)r=rt)rrN)rrrv) r1rrrrrrrzreduces_channelsr2s &&&&&&& r$r+ResNetBottleNeckLayer.__init__s  + ; J Jv{'4H]N;V Dcecncncp ]] 1OgVmn  ,Umvst u ,VZ [   !,r'cTpVPV4pVPV4pW, pVPV4pV#r9r}r~s&& r$r:ResNetBottleNeckLayer.forwardrr'r)r=r>Fr`rGs@@r$rrs--0r'rcXaa]tRt^toRtRV3RlV3RllltV3RlRltRtVtV;t #) ResNetStagez, A ResNet stage composed by stacked layers. c 8<V^8dQhRS[RS[RS[RS[RS[/#)rrLrrrdepth)rr)r"r#s"r$r%ResNetStage.__annotate__s=          r'c <\S V`4VPR8Xd\M\pVPR8Xd#V!VVVVP VP R7pMV!W#WAP R7p\P!V.\V^, 4Uu.uFq!W3VP R7NK upO5!Vn R#uupi) bottleneck)rrr)rrruN) r*r+ layer_typerrprRrrrxrangelayers) r1rLrrrrry first_layer_r2s &&&&&& r$r+ResNetStage.__init__s )/):):l)J%P`    ,!,,)/)H)H K &UfUfgKmm  dijorsjsdtudt_`5HYHYZdtu  usC c&<V^8dQhRS[RS[/#rirZ)r"r#s"r$r%rsVr'c@TpVPF pV!V4pK V#r9r)r1rjrmrys&& r$r:ResNetStage.forwards% [[E .L!r'r)rrr`rGs@@r$rrs#  4r'rcTaa]tRt^toV3RlV3RlltRV3RlRlltRtVtV;t#) ResNetEncoderc <V^8dQhRS[/#rKr)r"r#s"r$r%ResNetEncoder.__annotate__s\\|\r'c <\SV`4\P!.4VnVPP \ TVPVP^,VP'd^M^VP^,R74\VPVPR,4p\W!PR,4F.wwr4pVPP \ WWER74K0 R#))rrr=NN)rN) r*r+r ModuleListstagesappendrrQ hidden_sizesdownsample_in_first_stagedepthszip)r1rLin_out_channelsrrrr2s&& r$r+ResNetEncoder.__init__s mmB'   %%##A&"<< UY ' r'cV'dRMRpVPFpV'd WA3,pV!V4pK V'd WA3,pV'g/\;QJd.RW34FNK 5#!RW344#\VVR7#)Nc30"TF qfKVxK R#5ir9).0vs& r$ (ResNetEncoder.forward..sS$Aq$As )last_hidden_stater5r)rrr )r1rmrrr5 stage_modules&&&& r$r:ResNetEncoder.forwards3  KKL# - ? ' 5L ( )O;M5S\$AS5 S5S\$ASS S-*'  r')r)FTr?rGs@@r$rrs \\"   r'rczaa]tRtRtoRtRtR tRR.t]P!4V3Rl4t V3Rlt Rt Vt V;t#) ResNetPreTrainedModeliresnetrYrrcc<\SV`V4\V\P4'd&\ P !VPRRR7R#\V\P4'd\ P!VP\P!^4R7VPe\PP PVP4wr#V^8d^\P!V4, M^p\ P!VPV)V4R#R#RVP P"9d\ P$!VP4\ P&!VP4\ P&!VP(4\ P$!VP*4\-VRR4e#\ P&!VP.4R#R#R#)fan_outr>)mode nonlinearity)aN BatchNormnum_batches_tracked)r* _init_weights isinstancerr,initkaiming_normal_weightLinearkaiming_uniform_mathsqrtrr7_calculate_fan_in_and_fan_outuniform_r2r@ones_zeros_ running_mean running_vargetattrr)r1modulefan_inrboundr2s&& r$r#ResNetPreTrainedModel._init_weights sI f% fbii ( (  YV T  * *  ! !&--499Q< @{{&!HHMMGG V 17!DIIf-- fkkE659' F,,55 5 JJv}} % KK $ KK++ , JJv)) *v4d;G F667H 6r'c&<V^8dQh/S[;R&#rKr)r"r#s"r$r%"ResNetPreTrainedModel.__annotate__s r'r)image)r@rArBrCbase_model_prefixmain_input_nameinput_modalities_no_split_modulesr7no_gradr__annotate_func__rDrErFrGs@@r$rrsC!$O!*,<= ]]_88r'rcRaa]tRtRtoV3Rlt]RV3RlRll4tRtVtV;t #) ResNetModeli"c<\SV`V4Wn\V4Vn\ V4Vn\P!R4Vn VP4R#)r=N)r=r=) r*r+rLrIrSrencoderrAdaptiveAvgPool2drU post_initrVs&&r$r+ResNetModel.__init__$sI   (0 $V, **62  r'cN<V^8dQhRS[RS[R,RS[R,RS[/#rrYrNrr6)rr r )r"r#s"r$r%ResNetModel.__annotate__.s;   #Tk D[  2  r'c PVeTMVPPpVeTMVPPpVPV4pVP WRVR7pV^,pVP V4pV'gWx3VR,,#\ VVVPR7#)Nrrr)r pooler_outputr5)rLrrrSrrUr r5) r1rYrrkwargsembedding_outputencoder_outputsr pooled_outputs &&&&, r$r:ResNetModel.forward-s%9$D $++JjJj &1% 1` a classification loss is computed (Cross-Entropy). Nr:rNN)losslogitsr5)rLrrrr loss_functionr r5) r1rYrrrroutputsrrroutputs &&&&&, r$r:$ResNetForImageClassification.forwardas&1%>> from transformers import AutoImageProcessor, AutoBackbone >>> import torch >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> processor = AutoImageProcessor.from_pretrained("microsoft/resnet-50") >>> model = AutoBackbone.from_pretrained( ... "microsoft/resnet-50", out_features=["stage1", "stage2", "stage3", "stage4"] ... ) >>> inputs = processor(image, return_tensors="pt") >>> outputs = model(**inputs) >>> feature_maps = outputs.feature_maps >>> list(feature_maps[-1].shape) [1, 2048, 7, 7] ```NTr) feature_mapsr5 attentionsr) rLrrrSrr5 enumerate stage_names out_featuresr) r1rYrrrrr r5ridxstager s &&&&, r$r:ResNetBackbone.forwardsH&1%r*sP &!H .,-.   H %biiBryy0RYY"ryy4'BII'T#"))#L& BII& R8O88:( '( ( V ,s#8,s  ,s^ K ]$9K  K \ er'