a PSic@sVddlmZddlmZmZmZddlmZddlmZddl m Z m Z ddl m Z mZmZddlmZdd lmZd d lmZmZd d lmZd d lmZmZddlmZmZmZgdZ Gddde Z!GdddeZ"GdddeZ#edddfdddddeee#efe$e$ee"dd d!Z%d d"lm&Z&d d#l'm(Z(e&d$e#j)j*iZ+dS)%)partial)AnyOptionalUnion)Tensor)nn) QuantStub DeQuantStub)InvertedResidual MobileNetV2MobileNet_V2_Weights)Conv2dNormActivation)ImageClassification) WeightsEnumWeights)_IMAGENET_CATEGORIES)handle_legacy_interface_ovewrite_named_param) _fuse_modules _replace_reluquantize_model)QuantizableMobileNetV2MobileNet_V2_QuantizedWeights mobilenet_v2csLeZdZeeddfdd ZeedddZd eeddd d Z Z S) QuantizableInvertedResidualNargskwargsreturncs"tj|i|tj|_dSN)super__init__r quantizedFloatFunctionalskip_addselfrr  __class__g/var/www/html/django/DPS/env/lib/python3.9/site-packages/torchvision/models/quantization/mobilenetv2.pyr$sz$QuantizableInvertedResidual.__init__xr!cCs(|jr|j|||S||SdSr")use_res_connectr'addconvr)r/r,r,r-forwardsz#QuantizableInvertedResidual.forwardis_qatr!cCsNtt|jD]:}t|j|tjurt|jt|t|dg|ddqdS)NrTinplace)rangelenr2typerConv2drstr)r)r6idxr,r,r- fuse_model#sz&QuantizableInvertedResidual.fuse_model)N __name__ __module__ __qualname__rr$rr4rboolr? __classcell__r,r,r*r-rsrcsLeZdZeeddfdd ZeedddZd eeddd d Z Z S) rNrcs&tj|i|t|_t|_dS)zq MobileNet V2 main class Args: Inherits args from floating point MobileNetV2 N)r#r$rquantr dequantr(r*r,r-r$*szQuantizableMobileNetV2.__init__r.cCs"||}||}||}|Sr")rF _forward_implrGr3r,r,r-r45s   zQuantizableMobileNetV2.forwardr5cCsH|D]:}t|tur,t|gd|ddt|tur||qdS)N)012Tr7)modulesr;rrrr?)r)r6mr,r,r-r?;s    z!QuantizableMobileNetV2.fuse_model)Nr@r,r,r*r-r)s rc @sDeZdZedeeddddeddejdd d d id d dZ e Z dS)rzOhttps://download.pytorch.org/models/quantized/mobilenet_v2_qnnpack_37f702c5.pth) crop_sizeiz5)rrqnnpackzUhttps://github.com/pytorch/vision/tree/main/references/classification#qat-mobilenetv2z ImageNet-1Kg'1Q@gV@)zacc@1zacc@5z These weights were produced by doing Quantization Aware Training (eager mode) on top of the unquantized weights listed below. ) num_paramsmin_size categoriesbackendrecipe unquantized_metrics_docs)url transformsmetaN) rArBrCrrrrr IMAGENET1K_V1IMAGENET1K_QNNPACK_V1DEFAULTr,r,r,r-rCs$ r pretrainedcCs|ddrtjStjS)NquantizeF)getrr]r r\)r r,r,r-`s rb)weightsNTF)rcprogressr`)rcrdr`r r!cKs|rtnt|}|durLt|dt|jdd|jvrLt|d|jd|dd}tfdti|}t ||rt |||dur| |j |d|S)a Constructs a MobileNetV2 architecture from `MobileNetV2: Inverted Residuals and Linear Bottlenecks `_. .. note:: Note that ``quantize = True`` returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported. Args: weights (:class:`~torchvision.models.quantization.MobileNet_V2_QuantizedWeights` or :class:`~torchvision.models.MobileNet_V2_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.MobileNet_V2_QuantizedWeights` 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. quantize (bool, optional): If True, returns a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableMobileNetV2`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.MobileNet_V2_QuantizedWeights :members: .. autoclass:: torchvision.models.MobileNet_V2_Weights :members: :noindex: N num_classesrSrTrPblock)rd) rr verifyrr:r[poprrrrload_state_dictget_state_dict)rcrdr`r rTmodelr,r,r-r]s+   r) _ModelURLs) model_urlsZmobilenet_v2_qnnpack), functoolsrtypingrrrtorchrrtorch.ao.quantizationrr Ztorchvision.models.mobilenetv2r r r Zops.miscrZtransforms._presetsr_apirr_metar_utilsrrutilsrrr__all__rrrrDrrl mobilenetv2rmr]rYquant_model_urlsr,r,r,r-sH       7