g]dZddlZddlmZmZmZddlZddlZddlmZddl m Z m Z m Z ddl mZddlmZmZmZdd lmZdd lmZmZmZmZd d lmZejeZd ZdZ gdZ!dZ"dZ#dZ$dZ%dede&fdZ'd2dee&efde(fdZ)Gddej*Z+Gddej,Z-Gddej*Z.Gd d!ej*Z/Gd"d#ej*Z0Gd$d%ej*Z1Gd&d'ej*Z2Gd(d)ej*Z3Gd*d+eZ4ed,e$Gd-d.e4Z5ed/e$Gd0d1e4Z6dS)3zPyTorch EfficientNet model.N)OptionalTupleUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)BaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)add_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardlogging)EfficientNetConfigrzgoogle/efficientnet-b7)rirztabby, tabby cataN This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`EfficientNetConfig`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. aB Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Pixel values can be obtained using [`AutoImageProcessor`]. See [`AutoImageProcessor.__call__`] for details. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. config num_channelsc|j}||jz}t|t||dz z|z|z}|d|zkr||z }t|S)z< Round number of filters based on depth multiplier. g?) depth_divisorwidth_coefficientmaxint)rrdivisornew_dims r/var/www/html/ai-engine/env/lib/python3.11/site-packages/transformers/models/efficientnet/modeling_efficientnet.py round_filtersr"Psk"GF,,L'3|gk9::gEOPPG|###7 w<<T kernel_sizeadjustct|tr||f}|ddz|ddzf}|r$|ddz |d|ddz |dfS|d|d|d|dfS)aJ Utility function to get the tuple padding value for the depthwise convolution. Args: kernel_size (`int` or `tuple`): Kernel size of the convolution layers. adjust (`bool`, *optional*, defaults to `True`): Adjusts padding value to apply to right and bottom sides of the input. rrr) isinstancer)r$r%corrects r! correct_padr)_s+s##1"K0 1~"KNa$78G @ Q GAJNGAJGG GAJ GAJ??r#cLeZdZdZdeffd ZdejdejfdZxZ S)EfficientNetEmbeddingszL A module that corresponds to the stem module of the original work. rc|tt|d|_t jd|_t j|j|jdddd|_ t j |j|j |j |_ t|j|_dS) N )rrrrpaddingr rvalidFr$strider/bias)epsmomentum)super__init__r"out_dimr ZeroPad2dr/Conv2dr convolution BatchNorm2dbatch_norm_epsbatch_norm_momentum batchnormr hidden_act activationselfr __class__s r!r7zEfficientNetEmbeddings.__init__xs $VR00 |L999 9  1QPW^c    &:OZ`Ztuuu !23r# pixel_valuesreturnc||}||}||}||}|SN)r/r;r?rA)rCrEfeaturess r!forwardzEfficientNetEmbeddings.forwardsM<< --##H-->>(++??8,,r#) __name__ __module__ __qualname____doc__rr7torchTensorrJ __classcell__rDs@r!r+r+ssu 41 4 4 4 4 4 4ELU\r#r+c.eZdZ dfd ZxZS)EfficientNetDepthwiseConv2drr rTzerosc f||z} t|| ||||||| dS)N) in_channels out_channelsr$r2r/dilationgroupsr3 padding_mode)r6r7) rCrWdepth_multiplierr$r2r/rYr3r[rXrDs r!r7z$EfficientNetDepthwiseConv2d.__init__sV#%55  #%#%  r#)rr rrrTrU)rKrLrMr7rQrRs@r!rTrTsT           r#rTcXeZdZdZdedededeffd Zdejdej fd Z xZ S) EfficientNetExpansionLayerz_ This corresponds to the expansion phase of each block in the original implementation. rin_dimr8r2cttj||ddd|_tj||j|_t|j |_ dS)NrsameFrWrXr$r/r3) num_featuresr4) r6r7rr: expand_convr<r= expand_bnr r@ expand_act)rCrr_r8r2rDs r!r7z#EfficientNetExpansionLayer.__init__so 9     W&BWXXX !23r# hidden_statesrFc||}||}||}|SrH)rdrerfrCrgs r!rJz"EfficientNetExpansionLayer.forwards=((77 }55  66 r#) rKrLrMrNrrr7rO FloatTensorrPrJrQrRs@r!r^r^s 41 43 4 4VY 4 4 4 4 4 4U%65<r#r^c \eZdZdZdededededef fd Zdej d ej fd Z xZ S) EfficientNetDepthwiseLayerzk This corresponds to the depthwise convolution phase of each block in the original implementation. rr_r2r$adjust_paddingcvt||_|jdkrdnd}t||}t j||_t||||d|_t j ||j |j |_ t|j|_dS) Nrr0ra)r%r.Fr1rcr4r5)r6r7r2r)rr9depthwise_conv_padrTdepthwise_convr<r=r>depthwise_normr r@ depthwise_act) rCrr_r2r$rmconv_padr/rDs r!r7z#EfficientNetDepthwiseLayer.__init__s  "kQ..77Fk.AAA"$,w"?"?"?9  FHSX   !nV%:VE_   $F$56r#rgrFc|jdkr||}||}||}||}|S)Nr)r2rprqrrrsris r!rJz"EfficientNetDepthwiseLayer.forwardsa ;!   33MBBM++M:: ++M:: **=99 r# rKrLrMrNrrboolr7rOrjrPrJrQrRs@r!rlrls7"77 7  7  777777, U%6 5<        r#rlc ZeZdZdZd dedededeffd Zdej d ej fd Z xZ S) EfficientNetSqueezeExciteLayerzl This corresponds to the Squeeze and Excitement phase of each block in the original implementation. Frr_ expand_dimexpandct|r|n||_tdt ||jz|_tjd|_ tj |j|jdd|_ tj |j|jdd|_ t|j|_tj|_dS)Nr) output_sizera)rWrXr$r/)r6r7dimrrsqueeze_expansion_ratiodim_serAdaptiveAvgPool2dsqueezer:reducer{r r@ act_reduceSigmoid act_expand)rCrr_rzr{rDs r!r7z'EfficientNetSqueezeExciteLayer.__init__s !'3::V!S&*H!HIIJJ +::: i     i      !!23*,,r#rgrFc|}||}||}||}||}||}t j||}|SrH)rrrr{rrOmul)rCrginputss r!rJz&EfficientNetSqueezeExciteLayer.forwardss ]33  M22  66  M22  66  &-88 r#)FrvrRs@r!ryrys''1'3'C'Y]''''''* U%6 5<        r#ryc neZdZdZdedededededef fd Zd e j d e j d e j fd Z xZ S) EfficientNetFinalBlockLayerz[ This corresponds to the final phase of each block in the original implementation. rr_r8r2 drop_rateid_skipc t|dko| |_tj||ddd|_tj||j|j|_ tj ||_ dS)NrraFrbrop) r6r7 apply_dropoutrr: project_convr<r=r> project_bnDropoutdropout)rCrr_r8r2rrrDs r!r7z$EfficientNetFinalBlockLayer.__init__s #q[8[I     . f&;fF`   zI... r# embeddingsrgrFc||}||}|jr||}||z}|SrH)rrrr)rCrrgs r!rJz#EfficientNetFinalBlockLayer.forward$sR))-88  66   7 LL77M)J6Mr#rKrLrMrNrrfloatrwr7rOrjrPrJrQrRs@r!rrs/(/25/@C/MP/]b/mq//////"%"3EDUZ_Zfr#rcleZdZdZdededededededed ed effd Zd e j d e j fdZ xZ S)EfficientNetBlocka This corresponds to the expansion and depthwise convolution phase of each block in the original implementation. Args: config ([`EfficientNetConfig`]): Model configuration class. in_dim (`int`): Number of input channels. out_dim (`int`): Number of output channels. stride (`int`): Stride size to be used in convolution layers. expand_ratio (`int`): Expand ratio to set the output dimensions for the expansion and squeeze-excite layers. kernel_size (`int`): Kernel size for the depthwise convolution layer. drop_rate (`float`): Dropout rate to be used in the final phase of each block. id_skip (`bool`): Whether to apply dropout and sum the final hidden states with the input embeddings during the final phase of each block. Set to `True` for the first block of each stage. adjust_padding (`bool`): Whether to apply padding to only right and bottom side of the input kernel before the depthwise convolution operation, set to `True` for inputs with odd input sizes. rr_r8r2 expand_ratior$rrrmc t||_|jdkrdnd|_||z} |jrt ||| ||_t ||jr| n|||| |_t||| |j|_ t||jr| n||||||_ dS)NrTF)rr_r8r2)rr_r2r$rm)rr_rzr{)rr_r8r2rr) r6r7rr{r^ expansionrlrqrysqueeze_exciter projection) rCrr_r8r2rr$rrrm expand_in_dimrDs r!r7zEfficientNetBlock.__init__Js ("/144dd% - ; 7fmFDN9$(K;==V#)    =&]4;   6$(K;==V    r#rgrFc|}|jdkr||}||}||}|||}|S)Nr)rrrqrr)rCrgrs r!rJzEfficientNetBlock.forwardssg"   ! ! NN=99M++M:: ++M::  MBB r#rrRs@r!rr/s4' "' '  '  '  ' ' ' ' ' ' ' ' ' ' R U%6 5<        r#rc heZdZdZdeffd Z d dejdee dee d e fd Z xZ S) EfficientNetEncoderz Forward propogates the embeddings through each EfficientNet block. Args: config ([`EfficientNetConfig`]): Model configuration class. rc t|_|j_fdt |j}t fd|jD}d}g}t|D]}t||j|}t||j |}|j |} |j |} |j |} t|j|D]l} | dkrdnd} | dkrdn| } | dkr|n|}||jvrdnd}|j|z|z }t!|||| | | || | }|||dz }mt%j|_t%j|t|ddd d _t%j|j|j|j _t8|j_dS) NcVttjj|zSrH)rmathceildepth_coefficient)repeatsrCs r! round_repeatsz3EfficientNetEncoder.__init__..round_repeatss#ty!7'!ABBCC Cr#c3.K|]}|VdSrH).0nrs r! z/EfficientNetEncoder.__init__..s-LLaq))LLLLLLr#rTFr) rr_r8r2r$rrrrmirarbro)r6r7rrlenrWsumnum_block_repeatsranger"rXstrides kernel_sizes expand_ratiosdepthwise_paddingdrop_connect_raterappendr ModuleListblocksr:top_convr< hidden_dimr=r>top_bnr r@top_activation)rCrnum_base_blocks num_blockscurr_block_numrir_r8r2r$rjrrmrblockrrDs` @r!r7zEfficientNetEncoder.__init__sW  !'!9 D D D D Df011LLLL63KLLLLL '' $ $A"66+=a+@AAF#FF,?,BCCG^A&F -a0K!/2L==)A!)DEEFF $ $"#q&&$$e!ee$%EEv*8F.s"XXq!-----XXr#)last_hidden_staterg)rrrrtupler )rCrgrrall_hidden_statesrs r!rJzEfficientNetEncoder.forwards 1EN],,$[ 6 6E!E-00M# 6!m%55! m44  M22 ++M::  YXX]4E$FXXXXX X-++    r#)FT) rKrLrMrNrr7rOrjrrwr rJrQrRs@r!rrs581585858585858t05&*   ( 'tn d^  (         r#rc(eZdZdZeZdZdZgZdZ dS)EfficientNetPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. efficientnetrEct|tjtjfrT|jjd|jj|j |j j dSdSt|tj r?|j j |jj ddSdS)zInitialize the weightsg)meanstdNg?) r'rLinearr:weightdatanormal_rinitializer_ranger3zero_ LayerNormfill_)rCmodules r! _init_weightsz)EfficientNetPreTrainedModel._init_weightss fry")4 5 5 * M  & &CT[5R & S S S{&  &&((((('&  - - * K  " " $ $ $ M  $ $S ) ) ) ) ) * *r#N) rKrLrMrNr config_classbase_model_prefixmain_input_name_no_split_modulesrrr#r!rrsE &L&$O * * * * *r#rzUThe bare EfficientNet model outputting raw features without any specific head on top.c eZdZdeffd Zeeeee e de  d de j deedeed eee ffd ZxZS) EfficientNetModelrct|||_t||_t ||_|jdkr!tj |j d|_ nC|jdkr!tj |j d|_ ntd|j|dS)NrT) ceil_moderz2config.pooling must be one of ['mean', 'max'] got )r6r7rr+rrencoder pooling_typer AvgPool2drpooler MaxPool2d ValueErrorpooling post_initrBs r!r7zEfficientNetModel.__init__s     088*622   & ( (,v'8DIIIDKK  E ) ),v'8DIIIDKKbRXR`bbcc c r#vision) checkpoint output_typermodalityexpected_outputNrErrrFc||n |jj}||n |jj}|td||}||||}|d}||}||jdd}|s||f|ddzSt|||j S)Nz You have to specify pixel_valuesrrrrr)r pooler_outputrg) rruse_return_dictrrrrreshapeshaper rg)rCrErrembedding_outputencoder_outputsr pooled_outputs r!rJzEfficientNetModel.forwards%9$D $+Jj &1%> ? " " " JI" " " " " r#rz EfficientNet Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet. ceZdZfdZeeeeee e  d de j de e jde ede edeeeff d ZxZS) "EfficientNetForImageClassificationct||j|_||_t ||_t j|j|_ |jdkrt j |j |jnt j |_ |dS)Nrr)r6r7 num_labelsrrrrr dropout_raterrrIdentity classifierrrBs r!r7z+EfficientNetForImageClassification.__init__=s     + -f55zF$7888 KO?]^K^K^")F$5tGGGdfdodqdq r#)rrrrNrElabelsrrrFc||n |jj}||||}|r|jn|d}||}||}d}|Z|jjf|jdkr d|j_nN|jdkr7|jtj ks|jtj kr d|j_n d|j_|jjdkrWt} |jdkr1| | | }n| ||}n|jjdkrGt} | |d|j|d}n*|jjdkrt!} | ||}|s|f|ddz} ||f| zn| St#|||j S) 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). Nrr regressionsingle_label_classificationmulti_label_classificationr)losslogitsrg)rrrrrr problem_typerdtyperOlongrr rrviewrrrg) rCrErrroutputsrrrloss_fctoutputs r!rJz*EfficientNetForImageClassification.forwardIs (&1%r&s"! )))))))))) AAAAAAAAAA!!!!!! .----- ;:::::  H % %'/'31  ! , C    @@U3:.@@@@@(RY2     ")   44$$$$$$$$N$$$$$RY$$$N")BNNNNN NNNbW W W W W ")W W W t*****/***0[ < < < < < 3< < < ~! H H H H H )DH H H H H r#