j3<UddlZddlmZddlmZddlmZddlmZddl m Z ddl m Z ddl m Zdd l mZdd l mZdd lmZd d lmZmZgdZdddeeddddddeeddddddeeddddddeeddddddeeddddZeeeeeffed<dededededeejf d Z Gd!d"ej!Z d3d#ed$e"deed%eed&eed'ee"d(ee#eefdzd)eedzd*ede fd+Z$ d3d#ed$e"d)eedzd*ede f d,Z%d4d$e"d*ede fd.Z d4d$e"d*ede fd/Z&d4d$e"d*ede fd0Zd4d$e"d*ede fd1Z'd4d$e"d*ede fd2ZdS)5N)Callable)deepcopy)Any)nn) BasicBlock)ResNet)resnet18)resnet34)resnet50)VOCABS)conv_sequence_ptload_pretrained_params)rr resnet31r r resnet34_wide resnet_stage)gh|?5?g= ףp=?gV-?)gA`"?gl?g$C?)r rfrenchzKhttps://doctr-static.mindee.com/models?id=v0.4.1/resnet18-244bf390.pt&src=0)meanstd input_shapeclassesurlzKhttps://doctr-static.mindee.com/models?id=v0.4.1/resnet31-1056cc5c.pt&src=0zKhttps://doctr-static.mindee.com/models?id=v0.5.0/resnet34-bd8725db.pt&src=0zKhttps://doctr-static.mindee.com/models?id=v0.5.0/resnet50-1a6c155e.pt&src=0zPhttps://doctr-static.mindee.com/models?id=v0.6.0/resnet34_wide-b4b3e39e.pt&src=0)r rr r r default_cfgs in_channels out_channels num_blocksstridereturnc g}|}|}t|D]U}d}||kr!tjt||ddd|}|t |||||}d}V|S)zBuild a ResNet stageNFT) kernel_sizer)r downsample)ranger Sequentialrappendr) rrrr_layersin_chans_r#s m/var/www/html/Carbon-Document/venv/lib/python3.11/site-packages/doctr/models/classification/resnet/pytorch.pyrrAs!GGA :   l " "(8,PUW[ijst(u(u(uvJz'<jYYYZZZ  NceZdZdZ ddeedeedeed eed eeeefdzd ed ed eege j fdzdedede e e fdzddffd Zde de ddfdZxZS)raImplements a ResNet-31 architecture from `"Show, Attend and Read:A Simple and Strong Baseline for Irregular Text Recognition" `_. Args: num_blocks: number of resnet block in each stage output_channels: number of channels in each stage stage_conv: whether to add a conv_sequence after each stage stage_pooling: pooling to add after each stage (if None, no pooling) origin_stem: whether to use the orginal ResNet stem or ResNet-31's stem_channels: number of output channels of the stem convolutions attn_module: attention module to use in each stage include_top: whether the classifier head should be instantiated num_classes: number of output classes T@Nroutput_channels stage_stride stage_conv stage_pooling origin_stem stem_channels attn_module include_top num_classescfgrc |r0gtd|dddddtjddd} nFgtd|dzddddt|dz|ddddtjd} |g|dd z} t|| ||||D]\}}}}}}t ||||}|||||r(|t||dddd|'|tj|| tj|| rW| tjdtj dtj |d | d gtj | | |_ |D]}t|tjr(tj|jd d Dt|tjtjfrJtj|jdtj|jddS)Nr T)r"paddingrr!)r"rr=)r"r=)biasfan_outrelu)mode nonlinearityr)rr MaxPool2dziprr&extendr%AdaptiveAvgPool2dFlattenLinearsuper__init__r9modules isinstanceConv2dinitkaiming_normal_weight BatchNorm2d GroupNorm constant_r?)selfrr0r1r2r3r4r5r6r7r8r9r'in_chansn_blocksr(out_chanrconvpool_stagem __class__s r+rKzResNet.__init__ds  !!]D$AWXabccc 1a@@@GG !!]a%7tQR\]^^^!-1"4mT4]^hijjj QG "?_SbS%99?B /<]@ @ 3 3 ;Hgxt"'8XvFFF& kk(33444 j .x4[\fghhhiii bl400111 NN2=&1 2 2 2 2   NN$Q'' 1  /"-{FFF    '"" - -A!RY'' -''yv'VVVVA =>> -!!!(A...!!!&!,,,  - -r, path_or_urlkwargsc "t||fi|dSzLoad pretrained parameters onto the model Args: path_or_url: the path or URL to the model parameters (checkpoint) **kwargs: additional arguments to be passed to `doctr.models.utils.load_pretrained_params` NrrUr^r_s r+from_pretrainedzResNet.from_pretrained" t[;;F;;;;;r,)Tr.NTr/N)__name__ __module__ __qualname____doc__listintbooltuplerrModuledictstrrrKrd __classcell__)r]s@r+rrTsB  ,!9= %)6-6-I6-c6-3i 6- J 6- E#s(Od23 6-6-6-seRY./$66-6-6-#s(^d "6- 6-6-6-6-6-6-p<3<#<$<<<<<<<.from_pretrainedrer,rrv) rwrxrrryrprtypes MethodTyperdr9) rrrsarch_fnrtr_rzr{rdr|s r+ _tv_resnetrsZ#JJ}c,t:LY:W6X6XYYF= 9l4.@.KLLF9 L& ' 'D /DY'DO JJy G + +f + +d + + +E<3<#<$<<<<",_eDDES'-]&;s`_. >>> import torch >>> from doctr.models import resnet18 >>> model = resnet18(pretrained=False) >>> input_tensor = torch.rand((1, 3, 512, 512), dtype=torch.float32) >>> out = model(input_tensor) Args: pretrained: boolean, True if model is pretrained **kwargs: keyword arguments of the ResNet architecture Returns: A resnet18 model r rt fc.weightfc.bias)r tv_resnet18rsr_s r+r r ="   !),      r,c Ltd|gdgdgddgdzgdfdd d d gd |S) auResnet31 architecture with rectangular pooling windows as described in `"Show, Attend and Read:A Simple and Strong Baseline for Irregular Text Recognition", `_. Downsizing: (H, W) --> (H/8, W/4) >>> import torch >>> from doctr.models import resnet31 >>> model = resnet31(pretrained=False) >>> input_tensor = torch.rand((1, 3, 512, 512), dtype=torch.float32) >>> out = model(input_tensor) Args: pretrained: boolean, True if model is pretrained **kwargs: keyword arguments of the ResNet architecture Returns: A resnet31 model r)r!r<r )rr)r!r!r!r!T))r<r<)r<r!NNFz 13.weightz13.biasr4r5rtr}rs r+rr si$    $$$  ),      r,c 2td|tfdddgi|S)a ResNet-34 architecture as described in `"Deep Residual Learning for Image Recognition", `_. >>> import torch >>> from doctr.models import resnet34 >>> model = resnet34(pretrained=False) >>> input_tensor = torch.rand((1, 3, 512, 512), dtype=torch.float32) >>> out = model(input_tensor) Args: pretrained: boolean, True if model is pretrained **kwargs: keyword arguments of the ResNet architecture Returns: A resnet34 model r rtrr)r tv_resnet34rs r+r r *rr,c Ptd|gdgdgddgdzdgdzfdd d d gd |S) a<ResNet-34 architecture as described in `"Deep Residual Learning for Image Recognition", `_ with twice as many output channels. >>> import torch >>> from doctr.models import resnet34_wide >>> model = resnet34_wide(pretrained=False) >>> input_tensor = torch.rand((1, 3, 512, 512), dtype=torch.float32) >>> out = model(input_tensor) Args: pretrained: boolean, True if model is pretrained **kwargs: keyword arguments of the ResNet architecture Returns: A resnet34_wide model r)r rr )rrri)r!r<r<r<FrNTrz 10.weightz10.biasrrrs r+rrDsi"   !    ),      r,c 2td|tfdddgi|S)a ResNet-50 architecture as described in `"Deep Residual Learning for Image Recognition", `_. >>> import torch >>> from doctr.models import resnet50 >>> model = resnet50(pretrained=False) >>> input_tensor = torch.rand((1, 3, 512, 512), dtype=torch.float32) >>> out = model(input_tensor) Args: pretrained: boolean, True if model is pretrained **kwargs: keyword arguments of the ResNet architecture Returns: A resnet50 model r rtrr)r tv_resnet50rs r+r r drr,)N)F)(rcollections.abcrcopyrtypingrtorchrtorchvision.models.resnetrrTVResNetr rr rr rdoctr.datasetsr utilsrr__all__rjrrorp__annotations__rkrnrr%rlrmr}rrrrr,r+rsS $$$$$$000000888888==================!!!!!!======== e e e &$"4x())\ &$"4x())\ &$"4x())\ &$"4x())\ &$"4x())a ;$+$+ d3S#X&'$$$Nc#sW[\^\eWf&O<O<O<O<O