o ei0;@sDdZddlmZddlZddlmZddlmZddlmZddl m Z ddl m Z m Z mZmZdd lmZdd lmZmZd d lmZeeZGd ddejZGdddejZGdddejZGdddejZeGdddeZeGdddeZ eddGdddeZ!eddGddde eZ"gd Z#dS)!zPyTorch TextNet model.)AnyN)Tensor)ACT2CLS) BackboneMixin)BackboneOutputBaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)auto_docstringlogging) TextNetConfigcs8eZdZdeffdd ZdejdejfddZZS)TextNetConvLayerconfigcst|j|_|j|_|j|_t|jt r%|jdd|jddfn|jd}t j |j |j |j|j|dd|_t |j |j|_t |_|jdurVt|j|_dSdS)NrrF) kernel_sizestridepaddingbias)super__init__stem_kernel_sizer stem_strider stem_act_funcactivation_function isinstancetuplennConv2dstem_num_channelsstem_out_channelsconv BatchNorm2dbatch_norm_eps batch_normIdentity activationr)selfrr __class__j/home/ubuntu/transcripts/venv/lib/python3.10/site-packages/transformers/models/textnet/modeling_textnet.pyr's*    zTextNetConvLayer.__init__ hidden_statesreturncCs||}||}||SN)r#r&r()r)r.r,r,r-forwardBs   zTextNetConvLayer.forward) __name__ __module__ __qualname__rrtorchrr1 __classcell__r,r,r*r-r&src sLeZdZdZdededededef fdd Zd ejd ejfd d Z Z S) TextNetRepConvLayera This layer supports re-parameterization by combining multiple convolutional branches (e.g., main convolution, vertical, horizontal, and identity branches) during training. At inference time, these branches can be collapsed into a single convolution for efficiency, as per the re-parameterization paradigm. The "Rep" in the name stands for "re-parameterization" (introduced by RepVGG). r in_channels out_channelsrrc sft||_||_||_||_|ddd|dddf}t|_tj |||||dd|_ tj ||j d|_ |ddddf}d|dddf}|ddkrotj |||ddf||dd|_tj ||j d|_nd\|_|_|ddkrtj ||d|df||dd|_tj ||j d|_nd\|_|_||kr|dkrtj ||j d|_dSd|_dS)NrrrF)r8r9rrrr) num_featuresepsNN)rr num_channelsr9rrrReLUrr main_convr$r%main_batch_norm vertical_convvertical_batch_normhorizontal_convhorizontal_batch_norm rbr_identity) r)rr8r9rrrvertical_paddinghorizontal_paddingr*r,r-rRsZ         zTextNetRepConvLayer.__init__r.r/cCs||}||}|jdur||}||}||}|jdur0||}||}||}|jdur>||}||}||Sr0)r?r@rArBrCrDrEr)r)r. main_outputsvertical_outputshorizontal_outputsid_outr,r,r-r1s           zTextNetRepConvLayer.forward) r2r3r4__doc__rintrr5rr1r6r,r,r*r-r7Hs" 9r7cs.eZdZdedeffdd ZddZZS) TextNetStagerdepthc st|j|}|j|}t|}|j|}|j|d}|g|g|d}|g|} g} t|| ||D] } | t|g| Rq7t | |_ dS)Nr) rrconv_layer_kernel_sizesconv_layer_strideslen hidden_sizeszipappendr7r ModuleListstage) r)rrOrr num_layersstage_in_channel_sizestage_out_channel_sizer8r9rW stage_configr*r,r-rs     zTextNetStage.__init__cCs|jD]}||}q|Sr0)rW)r) hidden_stateblockr,r,r-r1s  zTextNetStage.forward)r2r3r4rrMrr1r6r,r,r*r-rNsrNc sLeZdZdeffdd Z  d dejdedBdedBdefd d Z Z S) TextNetEncoderrcsFtg}t|j}t|D] }|t||qt||_ dSr0) rrrRrPrangerUrNrrVstages)r)rr` num_stagesstage_ixr*r,r-rs   zTextNetEncoder.__init__Nr\output_hidden_states return_dictr/cCsL|g}|jD] }||}||q|s |f}|r||fS|St||dS)N)last_hidden_stater.)r`rUr)r)r\rcrdr.rWoutputr,r,r-r1s   zTextNetEncoder.forwardr<) r2r3r4rrr5rboolrr1r6r,r,r*r-r^s r^c@seZdZUeed<dZdZdS)TextNetPreTrainedModelrtextnet pixel_valuesN)r2r3r4r__annotations__base_model_prefixmain_input_namer,r,r,r-rhs rhc s`eZdZfddZe  d dededBdedBdeee efeeBe Bfdd Z Z S) TextNetModelcs8t|t||_t||_td|_| dS)N)rr) rrrstemr^encoderrAdaptiveAvgPool2dpooler post_initr)rr*r,r-rs     zTextNetModel.__init__Nrjrcrdr/c Ks|dur|n|jj}|dur|n|jj}||}|j|||d}|d}||}|s;||f} |r9| |dfS| St|||rF|ddSddS)Nrcrdrr)re pooler_outputr.)ruse_return_dictrcrorprrr ) r)rjrcrdkwargsr\encoder_outputsre pooled_outputrfr,r,r-r1s&   zTextNetModel.forwardr<) r2r3r4rr rrgrrlistr r1r6r,r,r*r-rns rnz TextNet Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet. ) custom_introc s\eZdZfddZe    d dejdBdejdBdedBdedBde f d d Z Z S) TextNetForImageClassificationcs|t||j|_t||_td|_t|_ |jdkr)t |j d|jnt |_ t|j|j g|_|dS)N)rrr)rr num_labelsrnrirrqavg_poolFlattenflattenLinearrSr'fcrV classifierrsrtr*r,r-rs    ( z&TextNetForImageClassification.__init__Nrjlabelsrcrdr/c Ks|dur|n|jj}|j|||d}|d}|jD]}||}q||} d} |dur3||| |j} |sI| f|dd} | durG| f| S| St| | |jdS)a labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the image classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Examples: ```python >>> import torch >>> import httpx >>> from io import BytesIO >>> from transformers import TextNetForImageClassification, TextNetImageProcessor >>> from PIL import Image >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> processor = TextNetImageProcessor.from_pretrained("czczup/textnet-base") >>> model = TextNetForImageClassification.from_pretrained("czczup/textnet-base") >>> inputs = processor(images=image, return_tensors="pt") >>> with torch.no_grad(): ... outputs = model(**inputs) >>> outputs.logits.shape torch.Size([1, 2]) ```Nrurr)losslogitsr.)rrwrirr loss_functionr r.) r)rjrrcrdrxoutputsrelayerrrrfr,r,r-r1s$   z%TextNetForImageClassification.forward)NNNN) r2r3r4rr r5 FloatTensor LongTensorrgr r1r6r,r,r*r-r}s$ r}zP TextNet backbone, to be used with frameworks like DETR and MaskFormer. c sTeZdZdZfddZe  d dededBdedBdeee Bfd d Z Z S) TextNetBackboneFcs*t|t||_|j|_|dSr0)rrrnrirSr:rsrtr*r,r-r\s   zTextNetBackbone.__init__Nrjrcrdr/c Ks|dur|n|jj}|dur|n|jj}|j|d|d}|r!|jn|d}d}t|jD]\}} | |jvr<|||f7}q,|sT|f} |rR|rI|jn|d}| |f7} | St||r^|jddSdddS)a Examples: ```python >>> import torch >>> import httpx >>> from io import BytesIO >>> from PIL import Image >>> from transformers import AutoImageProcessor, AutoBackbone >>> 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("czczup/textnet-base") >>> model = AutoBackbone.from_pretrained("czczup/textnet-base") >>> inputs = processor(image, return_tensors="pt") >>> with torch.no_grad(): >>> outputs = model(**inputs) ```NTrurr,) feature_mapsr. attentions) rrwrcrir. enumerate stage_names out_featuresr) r)rjrcrdrxrr.ridxrWrfr,r,r-r1es0  zTextNetBackbone.forwardr<) r2r3r4has_attentionsrr rrgrrr1r6r,r,r*r-rTs   r)rrnrhr})$rLtypingrr5torch.nnrr activationsrbackbone_utilsrmodeling_outputsrrr r modeling_utilsr utilsr r configuration_textnetr get_loggerr2loggerModulerr7rNr^rhrnr}r__all__r,r,r,r-s:        "Z)F E