fTh+SrSSKJr SSKJrJr SSKrSSKJr SSKJ r SSK J r J r SS K Jr S S KJr \"S S \ 55r\ "SS\ 55r"SS\R&5r"SS\R&5r"SS\R&5r"SS\R&5r"SS\R&5r\ "SS9"SS\55rSS/rg)zPyTorch ViTMatte model.) dataclass)OptionalTupleN)nn)PreTrainedModel) ModelOutputauto_docstring) load_backbone)VitMatteConfigc\rSrSr%SrSr\\R\ S'Sr \\R\ S'Sr \\ \R\ S'Sr \\ \R\ S'Srg) ImageMattingOutputa{ Class for outputs of image matting models. Args: loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Loss. alphas (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Estimated alpha values. hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states (also called feature maps) of the model at the output of each stage. attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, patch_size, sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads. Nlossalphas hidden_states attentions)__name__ __module__ __qualname____firstlineno____doc__rrtorch FloatTensor__annotations__rrrr__static_attributes__rf/var/www/auris/envauris/lib/python3.13/site-packages/transformers/models/vitmatte/modeling_vitmatte.pyrrsg()-D(5$$ %,*.FHU&& '.8[TU]5 [R"UUSUUSS9Ul[R "X1R S9Ul[R"5Ul g)NrF) in_channels out_channels kernel_sizestridepaddingr/)eps) super__init__rr)conv BatchNorm2dbatch_norm_eps batch_normReLUrelu)r1r-r=r>r@rA __class__s r rDVitMatteBasicConv3x3.__init__LsU II#%   ..;P;PQGGI rclURU5nURU5nURU5nU$N)rErHrJr1 hidden_states r forwardVitMatteBasicConv3x3.forwardYs2yy. |4 yy. r)rHrErJ)r rrrrrrDrQr __classcell__rKs@r r:r:Gs rr:c2^\rSrSrSrU4SjrSrSrU=r$)VitMatteConvStreamaz[ Simple ConvStream containing a series of basic conv3x3 layers to extract detail features. c>[TU]5 SnURbURRnURn[ R "5UlU/U-Ul[[UR5S- 5HInURUnURUS-nURR[XU55 MK g)Nr ) rCrDbackbone_config num_channelsconvstream_hidden_sizesr ModuleListconvs conv_chansrangelenappendr:)r1r-r=r>iin_chan_ out_chan_rKs r rDVitMatteConvStream.__init__fs   ! ! - 00==K55 ]]_ &-,6s4??+a/0Aq)HA.I JJ  26YO P1rcSU0nUn[[UR55H-nURU"U5nS[US-5-nX2U'M/ U$)Ndetailed_feature_map_0detailed_feature_map_r )rbrcr`str)r1r$out_dict embeddingsrename_s r rQVitMatteConvStream.forwardysZ,l;! s4::'AAz2J+c!a%j8E(UO( r)rar`rTrVs@r rXrXasQ&rrXc2^\rSrSrSrU4SjrSrSrU=r$)VitMatteFusionBlockzT Simple fusion block to fuse features from ConvStream and Plain Vision Transformer. cD>[TU]5 [XUSSS9Ulg)Nr )r@rA)rCrDr:rE)r1r-r=r>rKs r rDVitMatteFusionBlock.__init__s" (lST^_` rc[RRUSSSS9n[R"X#/SS9nUR U5nU$)NrSbilinearF) scale_factormode align_cornersr )dim)r functional interpolatercatrE)r1featuresdetailed_feature_mapupscaled_featuresouts r rQVitMatteFusionBlock.forwardsHMM55hQU_ot5uii-AqIiin r)rErTrVs@r rrrrsarrrc2^\rSrSrSrU4SjrSrSrU=r$) VitMatteHeadzB Simple Matting Head, containing only conv3x3 and conv1x1 layers. c &>[TU]5 URSnSn[R"[R "X#SSSS9[R "U5[R"S5[R "USSSSS95Ulg)Nrr )r?r@rATr) rCrDfusion_hidden_sizesr Sequentialr)rFrI matting_convs)r1r-r= mid_channelsrKs r rDVitMatteHead.__init__sr 004  ]] IIkQqRS T NN< ( GGDM IIlA1Q J  rc(URU5nU$rNrrOs r rQVitMatteHead.forwards)),7 rrrTrVs@r rrs  rrc2^\rSrSrSrU4SjrSrSrU=r$)VitMatteDetailCaptureModulez? Simple and lightweight Detail Capture Module for ViT Matting. c >[TU]5 [UR5[UR5S-:wa [ S5eXl[U5UlURRUl [R"5Ul UR/UR-Ul[[UR5S- 5HXnURR![#UURUURUS-*-URUS-S95 MZ [%U5Ulg)Nr z_The length of fusion_hidden_sizes should be equal to the length of convstream_hidden_sizes + 1.)r-r=r>)rCrDrcrr^ ValueErrorr-rX convstreamrarr_ fusion_blocks hidden_sizefusion_channelsrbrdrrr matting_head)r1r-rerKs r rD$VitMatteDetailCaptureModule.__init__s  v)) *c&2P2P.QTU.U Uq  ,V4//44]]_ & 2 23f6P6PPs4//0145A    % %#! $ 4 4Q 7$//APQE(:S S!%!5!5a!e!< 6)0rc:URU5n[[UR55HBnS[ [UR5U- S- 5-nURU"XU5nMD [ R "URU55nU$)Nrkr )rrbrcrrlrsigmoidr)r1rr$detail_featuresredetailed_feature_map_namers r rQ#VitMatteDetailCaptureModule.forwards//,7s4--./A(?#c$J\J\F]`aFadeFeBf(f %))!,XG`7abH0t00:; r)r-rarrrrrTrVs@r rrs12rrzX ViTMatte framework leveraging any vision backbone e.g. for ADE20k, CityScapes. ) custom_introc ^\rSrSrU4Sjr\S S\\RS\\ S\\ S\\RS\\ 4 Sjj5r S r U=r $) VitMatteForImageMattingc>[TU]U5 Xl[U5Ul[ U5UlUR5 grN)rCrDr-r backbonerdecoder post_init)r1r-rKs r rD VitMatteForImageMatting.__init__s9   %f- 26:  rr$output_attentionsoutput_hidden_stateslabels return_dictcUbUOURRnUbUOURRnUbUOURRnSnUb [ S5eUR R XUS9nURSnURX5n U(dU 4USS-n UbU4U -$U $[UU URURS9$)ap labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*): Ground truth image matting for computing the loss. Examples: ```python >>> from transformers import VitMatteImageProcessor, VitMatteForImageMatting >>> import torch >>> from PIL import Image >>> from huggingface_hub import hf_hub_download >>> processor = VitMatteImageProcessor.from_pretrained("hustvl/vitmatte-small-composition-1k") >>> model = VitMatteForImageMatting.from_pretrained("hustvl/vitmatte-small-composition-1k") >>> filepath = hf_hub_download( ... repo_id="hf-internal-testing/image-matting-fixtures", filename="image.png", repo_type="dataset" ... ) >>> image = Image.open(filepath).convert("RGB") >>> filepath = hf_hub_download( ... repo_id="hf-internal-testing/image-matting-fixtures", filename="trimap.png", repo_type="dataset" ... ) >>> trimap = Image.open(filepath).convert("L") >>> # prepare image + trimap for the model >>> inputs = processor(images=image, trimaps=trimap, return_tensors="pt") >>> with torch.no_grad(): ... alphas = model(**inputs).alphas >>> print(alphas.shape) torch.Size([1, 1, 640, 960]) ```NzTraining is not yet supported)rrrr )rrrr) r-use_return_dictrrNotImplementedErrorrforward_with_filtered_kwargs feature_mapsrrrr) r1r$rrrrroutputsrroutputs r rQVitMatteForImageMatting.forwardsR&1%B 'tn B & B d^ B B rr)r dataclassesrtypingrrrrmodeling_utilsrutilsr r utils.backbone_utilsr configuration_vitmatter rr"Moduler:rXrrrrr__all__rrr rs!" -012 :: :6 )o ) )2994  F"))"2990&"))&R N 5N  N b %&? @r