:jAddlZddlmZddlmZddlmZmZddlZddl m Z ddl m cm Z ddlmcmZddlmZddlmZddlmZd d lmZmZmZd d lmZd d lmZmZgd ZGdde j@Z!Gdde jDZ#Gdde jHZ%Gdde j@Z&de j@dede'ddfdZ(de)de*e)e)e)e)fde)deede'dede&fdZ+d ed!d"d#Z,Gd$d%eZ-Gd&d'eZ.Gd(d)eZ/Gd*d+eZ0eed,e-jbf-dd.d/dee-de'dede&fd0Z2eed,e.jbf-dd.d/dee.de'dede&fd1Z3eed,e/jbf-dd.d/dee/de'dede&fd2Z4eed,e0jbf-dd.d/dee0de'dede&fd3Z5y)4N) OrderedDict)partial)AnyOptional)Tensor)ImageClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) DenseNetDenseNet121_WeightsDenseNet161_WeightsDenseNet169_WeightsDenseNet201_Weights densenet121 densenet161 densenet169 densenet201c JeZdZ ddedededededdf fd Zd eedefd Z d eedefd Z e jjd eedefd Ze jjd eedefdZe jjd edefdZd edefdZxZS) _DenseLayernum_input_features growth_ratebn_size drop_ratememory_efficientreturnNct|tj||_tj d|_tj|||zddd|_tj||z|_ tj d|_ tj||z|dddd|_ t||_ ||_y)NTinplacer F kernel_sizestridebiasr'r(paddingr))super__init__nn BatchNorm2dnorm1ReLUrelu1Conv2dconv1norm2relu2conv2floatr r!)selfrrrr r! __class__s p/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torchvision/models/densenet.pyr.z_DenseLayer.__init__ s ^^$67 WWT* YY17[3HVW`ahmn ^^Gk$9: WWT* YYw4kqYZdelqr y) 0inputsctj|d}|j|j|j |}|SNr )torchcatr5r3r1)r:r>concated_featuresbottleneck_outputs r< bn_functionz_DenseLayer.bn_function/s;!IIfa0 JJtzz$**=N2O'PQ  r=inputc.|D]}|jsyy)NTF) requires_grad)r:rFtensors r<any_requires_gradz_DenseLayer.any_requires_grad5s" F## r=c@fd}tj|g|ddiS)Nc&j|SN)rE)r>r:s r<closurez7_DenseLayer.call_checkpoint_bottleneck..closure=s##F+ +r= use_reentrantF)cp checkpoint)r:rFrNs` r<call_checkpoint_bottleneckz&_DenseLayer.call_checkpoint_bottleneck;s# ,}}WBuBEBBr=cyrMr:rFs r<forwardz_DenseLayer.forwardB r=cyrMrTrUs r<rVz_DenseLayer.forwardFrWr=ct|tr|g}n|}|jrL|j|r;tj j r td|j|}n|j|}|j|j|j|}|jdkDr,tj||j|j }|S)Nz%Memory Efficient not supported in JITr)ptraining) isinstancerr!rJrAjit is_scripting ExceptionrRrEr8r7r6r Fdropoutr[)r:rF prev_featuresrD new_featuress r<rVz_DenseLayer.forwardLs eV $"GM!M  T%;%;M%Jyy%%' GHH $ ? ? N  $ 0 0 ? zz$**TZZ8I-J"KL >>A 99\T^^dmm\Lr=F)__name__ __module__ __qualname__intr9boolr.listrrErJrAr]unusedrR_overload_methodrV __classcell__r;s@r<rrs rw 1"% 147 1BE 1RW 1ko 1  1!$v,!6! tF|  YYCV CCC  YY T&\ f    YY V    Vr=rcTeZdZdZ d dededededededd ffd Zd edefd Z xZ S) _DenseBlockr num_layersrrrr r!r"Nct |t|D]0}t|||zz||||}|j d|dzz|2y)N)rrr r!z denselayer%dr )r-r.ranger add_module) r:rqrrrr r!ilayerr;s r<r.z_DenseBlock.__init__cs_ z" =A"Q_4'#!1 E OONa!e4e < =r= init_featuresc|g}|jD]\}}||}|j| tj|dSr@)itemsappendrArB)r:rwfeaturesnamervrcs r<rVz_DenseBlock.forwardwsJ!?::< *KD% ?L OOL ) *yy1%%r=rd) rerfrg_versionrhr9rir.rrVrmrns@r<rprp`siH"'== = =  =  == =(&V&&r=rpc,eZdZdededdffd ZxZS) _Transitionrnum_output_featuresr"Nct|tj||_tj d|_tj||ddd|_tjdd|_ y)NTr$r Fr&r)r'r() r-r.r/r0normr2relur4conv AvgPool2dpool)r:rrr;s r<r.z_Transition.__init__s^ NN#56 GGD) II02EST]^ejk LLQq9 r=)rerfrgrhr.rmrns@r<rrs":3:S:T::r=rcreZdZdZ ddedeeeeefdededededed d ffd Zd e d e fd Z xZ S)raKDensenet-BC model class, based on `"Densely Connected Convolutional Networks" `_. Args: growth_rate (int) - how many filters to add each layer (`k` in paper) block_config (list of 4 ints) - how many layers in each pooling block num_init_features (int) - the number of filters to learn in the first convolution layer bn_size (int) - multiplicative factor for number of bottle neck layers (i.e. bn_size * k features in the bottleneck layer) drop_rate (float) - dropout rate after each dense layer num_classes (int) - number of classification classes memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient, but slower. Default: *False*. See `"paper" `_. r block_confignum_init_featuresrr num_classesr!r"Nc "t|t|tjt dtj d|ddddfdtj|fdtjd fd tjddd fg|_ |}t|D]\} } t| |||||} |jjd| d zz| || |zz}| t|d z k7sSt||dz} |jjd| d zz| |dz}|jjdtj|tj |||_|j%D]} t'| tj r*tj(j+| j,Gt'| tjrUtj(j/| j,d tj(j/| j0dt'| tj stj(j/| j0dy)Nconv0r*rFr+norm0relu0Tr$pool0r )r'r(r,)rqrrrr r!z denseblock%d)rrz transition%dnorm5r)r-r.r r/ Sequentialrr4r0r2 MaxPool2dr{ enumeraterprtlenrLinear classifiermodulesr\initkaiming_normal_weight constant_r))r:rrrrr rr! num_featuresrurqblocktransmr;s r<r.zDenseNet.__init__s D! bii+?@bggd34bllqANO     ) &|4 1MAz%#/'#!1 E MM $ $^q1u%=u E'*{*BBLC %))#|YeijYjk ((1q5)A5I+q0  1"   "..*FG))L+> -A!RYY'''1Ar~~.!!!((A.!!!&&!,Aryy)!!!&&!, -r=xc|j|}tj|d}tj|d}t j |d}|j |}|S)NTr$)r r r )r{r`radaptive_avg_pool2drAflattenr)r:rr{outs r<rVzDenseNet.forwardsU==#ffXt,##C0mmC#ooc" r=)  @riF) rerfrg__doc__rhtupler9rir.rrVrmrns@r<rrs "2A!#!&:-:-Cc3./:- :-  :-  :-:-:- :-xFr=rmodelweightsprogressr"c6tjd}|j|d}t|j D]D}|j |}|s|j d|j dz}||||<||=F|j|y)Nz]^(.*denselayer\d+\.(?:norm|relu|conv))\.((?:[12])\.(?:weight|bias|running_mean|running_var))$T)r check_hashr r)recompileget_state_dictrjkeysmatchgroupload_state_dict)rrrpattern state_dictkeyresnew_keys r<_load_state_dictrs jjhG''d'KJJOO%& mmC  iilSYYq\1G",S/Jw 3   *%r=rrrkwargsc |#t|dt|jdt|||fi|}|t ||||S)Nr categories)rrr)rrmetarr)rrrrrrrs r< _densenetrsOfmSl9S5TU [,0A LV LEugI Lr=)rz*https://github.com/pytorch/vision/pull/116z'These weights are ported from LuaTorch.)min_sizerrecipe_docsc ReZdZedeediedddddid d d  ZeZy )rz@ num_params_metrics_ops _file_sizeurl transformsrN rerfrgr rr _COMMON_META IMAGENET1K_V1DEFAULTrTr=r<rr sQ J.#>    !##     M Gr=rc ReZdZedeediedddddid d d  ZeZy )rz    "##  !  M Gr=rc ReZdZedeediedddddid d d  ZeZy )rz    "##     M Gr=rc ReZdZedeediedddddid d d  ZeZy )rz    "##     M Gr=r pretrained)rT)rrc Ltj|}tddd||fi|S)a{Densenet-121 model from `Densely Connected Convolutional Networks `_. Args: weights (:class:`~torchvision.models.DenseNet121_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.DenseNet121_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.densenet.DenseNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.DenseNet121_Weights :members: rrr)rverifyrrrrs r<rr[,*"((1G R"gx J6 JJr=c Ltj|}tddd||fi|S)a{Densenet-161 model from `Densely Connected Convolutional Networks `_. Args: weights (:class:`~torchvision.models.DenseNet161_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.DenseNet161_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.densenet.DenseNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.DenseNet161_Weights :members: 0)rr$r`)rrrrs r<rrurr=c Ltj|}tddd||fi|S)a{Densenet-169 model from `Densely Connected Convolutional Networks `_. Args: weights (:class:`~torchvision.models.DenseNet169_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.DenseNet169_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.densenet.DenseNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.DenseNet169_Weights :members: r)rrrrr)rrrrs r<rrrr=c Ltj|}tddd||fi|S)a{Densenet-201 model from `Densely Connected Convolutional Networks `_. Args: weights (:class:`~torchvision.models.DenseNet201_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.DenseNet201_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.densenet.DenseNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.DenseNet201_Weights :members: r)rrrrr)rrrrs r<rrrr=)6r collectionsr functoolsrtypingrrrAtorch.nnr/torch.nn.functional functionalr`torch.utils.checkpointutilsrQrPrtransforms._presetsr r _apir r r_metar_utilsrr__all__Moduler ModuleDictrprrrrirrhrrrrrrrrrrrrrTr=r<rs # ##5'66'B >"))>B&"--&>:"--:RryyRj&BII& &t&PT&&S#s*+k "     (&: :  +(+(+(+(,0C0Q0Q!RS<@SWKH%89KDKcfKksKTK0,0C0Q0Q!RS<@SWKH%89KDKcfKksKTK0,0C0Q0Q!RS<@SWKH%89KDKcfKksKTK0,0C0Q0Q!RS<@SWKH%89KDKcfKksKTKr=