[TU]5 [R"UUUUUUS9Ul[R "U5Ul[R"5Ulg)N)rrrrrr) super__init__rConv2dconv BatchNorm2d batch_normReLU activation)selfrrrrrr __class__s d/var/www/html/shao/venv/lib/python3.13/site-packages/transformers/models/upernet/modeling_upernet.pyrUperNetConvModule.__init__$sQ II#%#   ..6'')inputclURU5nURU5nURU5nU$N)rrr )r!r&outputs r#forwardUperNetConvModule.forward9s15!(( r%)r rr)rFr )__name__ __module__ __qualname____firstlineno____doc__intrtuplestrboolrtorchTensorr*__static_attributes__ __classcell__r"s@r#rrs5601$$$3c3h/0 $ sE#s(OS01 $  $U38_,-$ $$*U\\ellr%rcv^\rSrSrS\S\S\SS4U4SjjrS\RS\R4S jrS r U=r $) UperNetPyramidPoolingBlockA pool_scalerchannelsrNc>[TU]5 [R"U5[ X#SS9/Ul[ UR 5H upEUR[U5U5 M" g)Nr r) rrrAdaptiveAvgPool2drlayers enumerate add_moduler3)r!r=rr>ilayerr"s r#r#UperNetPyramidPoolingBlock.__init__BsX    , k C  "$++.HA OOCFE */r%r&c@UnURH nU"U5nM U$r(rB)r!r& hidden_staterFs r#r*"UperNetPyramidPoolingBlock.forwardKs% [[E .L!r%rI) r,r-r.r/r1rr5r6r*r7r8r9s@r#r;r;AsD+3+S+C+D+U\\ellr%r;c ^\rSrSrSrS\\S4S\S\S\SS 4 U4S jjrS \ RS\ \ R4S jr S r U=r$)UperNetPyramidPoolingModuleRaQ Pyramid Pooling Module (PPM) used in PSPNet. Args: pool_scales (`tuple[int]`): Pooling scales used in Pooling Pyramid Module. in_channels (`int`): Input channels. channels (`int`): Channels after modules, before conv_seg. align_corners (`bool`): align_corners argument of F.interpolate. pool_scales.rr> align_cornersrNc>[TU]5 XlX@lX lX0l/Ul[U5HEupV[XbUS9nUR RU5 UR[U5U5 MG g)N)r=rr>) rrrOrPrr>blocksrCr;appendrDr3) r!rOrr>rPrEr=blockr"s r#r$UperNetPyramidPoolingModule.__init__asn &*&  &{3MA.*hpqE KK  u % OOCFE *4r%xc/nURHVnU"U5n[RRXAR 5SSSUR S9nUR U5 MX U$)NbilinearsizemoderP)rRr functional interpolater[rPrS)r!rVppm_outsppmppm_outupsampled_ppm_outs r#r*#UperNetPyramidPoolingModule.forwardmsg;;C!fG " 9 9ffhqrl4K]K]!:!  OO- .  r%)rPrRr>rrO)r,r-r.r/r0r2r1r4rr5r6listr*r7r8r9s@r#rMrMRs`  +E#s(O +# +QT +ei +nr +$u||*<r%rMcl^\rSrSrSrU4SjrSrS\RS\R4Sjr Sr U=r $) UperNetHeadxz Unified Perceptual Parsing for Scene Understanding. This head is the implementation of [UPerNet](https://huggingface.co/papers/1807.10221). c>[TU]5 XlURUlX lUR UlSUl[R"UR URSS9Ul [URURSUR URS9Ul [URS[UR5UR --UR SSS9Ul[R""5Ul[R""5UlURSSHmn[X R SS9n[UR UR SSS9nUR$R)U5 UR&R)U5 Mo [[UR5UR -UR SSS9Ulg)NFr r@)rPrrr)rrconfigrOr hidden_sizer>rPrr num_labels classifierrM psp_modulesrlen bottleneck ModuleList lateral_convs fpn_convsrSfpn_bottleneck)r!rkrl_convfpn_convr"s r#rUperNetHead.__init__~s  !--&** "))DMM63D3DRST7      R MM,,   ,   R 3t'7'7#84==#H H MM   ]]_++CR0K&{MMqQF( ST^_`H    % %f - NN ! !( + 1 0   !DMM 1 MM  r%cUSnU/nURURU55 [R"USS9nUR U5nU$)Nrir dim)extendror5catrq)r!inputsrVpsp_outsr)s r# psp_forwardUperNetHead.psp_forwardsL 2J3((+,99X1-* r%encoder_hidden_statesrc <[UR5VVs/sHup#U"X5PM nnnURURU55 [ U5n[ US- SS5HPnXBS- R SSnXBS- [RRXBUSURS9-XBS- 'MR [ US- 5Vs/sHo RU"XB5PM nnURUS5 [ US- SS5HAn[RRXrUSR SSSURS9Xr'MC [R"USS9nURU5nURU5nU$s snnfs snf)Nr rrirXrYrZrz)rCrsrSrrprangeshaperr]r^rPrtr5r}rurn) r!rrE lateral_convlateralsused_backbone_levels prev_shapefpn_outsr)s r#r*UperNetHead.forwardsR[\`\n\nRopRoqL!6!9:Rop(()>?@ #8}+a/B7A!a%..qr2J&1uo 0I0I *:TM_M_1J1HUO8=BBVYZBZ<[\<[qNN1%hk2<[\ %+a/B7A--33 (1+"3"3AB"7jX\XjXj4HK899X1-$$X.( 3q]s FF) rPrqr>rnrkrurtrrsrOro) r,r-r.r/r0rrr5r6r*r7r8r9s@r#rfrfxs3 % NU\\ellr%rfc ^\rSrSrSrS S\S\S\\\\\44SS4U4SjjjrS \ RS\ R4S jr S r U=r $) UperNetFCNHeada Fully Convolution Networks for Semantic Segmentation. This head is the implementation of [FCNNet](https://huggingface.co/papers/1411.4038>). Args: config: Configuration. in_channels (int): Number of input channels. kernel_size (int): The kernel size for convs in the head. Default: 3. dilation (int): The dilation rate for convs in the head. Default: 1. in_indexrrrNc `>[T U]5 XlURcX#O URUlUR UlURUlURUl X0l US-U-n/nUR[URUR XFUS95 [URS- 5H2nUR[UR UR XFUS95 M4 URS:Xa[R "5UlO[R$"U6UlUR(a4[URUR -UR XDS-S9Ul[R("UR UR*SS9Ulg)NrX)rrrr rrjr@)rrrkauxiliary_in_channelsrauxiliary_channelsr>auxiliary_num_convs num_convsauxiliary_concat_input concat_inputrrSrrrIdentityconvs Sequentialconv_catrrmrn) r!rkrrrr conv_paddingrrEr"s r#rUperNetFCNHead.__init__sb  %+%A%A%IK !vOkOk 11 33"99 #q(H4    $--[iq  t~~)*A LL!MM4==kjr + >>Q DJ.DJ   -  4==0$--[qrbrDM))DMM63D3DRSTr%rcXRnURU5nUR(a%UR[R "X#/SS95nUR U5nU$)Nr rz)rrrrr5r}rn)r!r hidden_statesr)s r#r*UperNetFCNHead.forwardsT-mm< M*   ]]599m-D!#LMF( r%) r>rnrrkrrrrr)rXrr )r,r-r.r/r0r1rr2rr5r6r*r7r8r9s@r#rrsx  uv$U-0$UCF$UV[\_afgjlogoap\pVq$U $U$ULU\\ellr%rc.\rSrSr%\\S'Sr/rSrSr g)UperNetPreTrainedModeli rk pixel_valuesc[U[R5(akURRR SUR RS9 URb%URRR5 gg[U[R5(aJURRRS5 URRR5 gg)Ng)meanstdg?) isinstancerrweightdatanormal_rkinitializer_rangerzero_rfill_)r!modules r# _init_weights$UperNetPreTrainedModel._init_weightss fbii ( ( MM   & &CT[[5R5R & S{{&   &&('  / / MM   $ $S ) KK   " " $0r%N) r,r-r.r/r __annotations__main_input_name_no_split_modulesrr7rr%r#rr s $O%r%rzW UperNet framework leveraging any vision backbone e.g. for ADE20k, CityScapes. ) custom_introc^\rSrSrU4Sjr\S S\\RS\\ S\\ S\\RS\\ S\ \ \ 44 S jj5r S rU=r$) UperNetForSemanticSegmentationic>[TU]U5 [U5Ul[ XRR S9UlUR(a[XRR S9OSUl UR5 g)N)r) rrr backbonerfr> decode_headuse_auxiliary_headrauxiliary_head post_init)r!rkr"s r#r'UperNetForSemanticSegmentation.__init__"sf  %f- 'v==;Q;QRJPJcJcN6}}/E/E Fim  r%routput_attentionsoutput_hidden_stateslabels return_dictrcUb%URRS:Xa [S5eUbUOURRnUbUOURRnUbUOURR nUR RXUS9nURnURU5n[RRXRSSSSS9nSn URb=URU5n [RRXRSSSSS9n Sn UbK[URR S 9n U "X5n U b#U "X5n XRR"U -- n U(d%U(a U4USS-n O U4USS-n U bU 4U -$U $[%U UUR&UR(S 9$) a( labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*): Ground truth semantic segmentation maps for computing the loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels > 1`, a classification loss is computed (Cross-Entropy). Examples: ```python >>> from transformers import AutoImageProcessor, UperNetForSemanticSegmentation >>> from PIL import Image >>> from huggingface_hub import hf_hub_download >>> image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-tiny") >>> model = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-tiny") >>> filepath = hf_hub_download( ... repo_id="hf-internal-testing/fixtures_ade20k", filename="ADE_val_00000001.jpg", repo_type="dataset" ... ) >>> image = Image.open(filepath).convert("RGB") >>> inputs = image_processor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits # shape (batch_size, num_labels, height, width) >>> list(logits.shape) [1, 150, 512, 512] ```Nr z/The number of labels should be greater than one)rrrXrYFrZ) ignore_index)losslogitsr attentions)rkrm ValueErroruse_return_dictrrrforward_with_filtered_kwargs feature_mapsrrr]r^rrrloss_ignore_indexauxiliary_loss_weightrrr)r!rrrrroutputsfeaturesrauxiliary_logitsrloss_fctauxiliary_lossr)s r#r*&UperNetForSemanticSegmentation.forward0sH  $++"8"8A"=NO O%0%P 'tn P & P d^ P  u-- .P P r%r)r0typingrrr5rtorch.nnrmodeling_outputsrmodeling_utilsr utilsr utils.backbone_utilsr configuration_upernetr Modulerr;rMrfrrr__all__rr%r#rsy" %7-#10  F"#"))#LQ"))Qh=RYY=@ %_ % % ` %;`  ` F ,-E Fr%