o )i?* @sddlZddlmZddlmZmZmZmZddlZddl m Z ddl m m Z ddlmZddlmZddlmZmZddlmZd d lmZmZmZd d lmZd d lmZmZd dl m!Z!m"Z"m#Z#gdZ$Gdddej%Z&Gdddej'Z(Gdddej)Z*Gdddej+Z,Gdddej-Z.Gdddej/Z0Gdddej1Z2Gdddej3Z4Gd d!d!eZ5ed"d#ed$d%d&fd'dd(d)d*d+eee5efd,e6d-e6d.ed/e4f d0d1Z7dS)2N)partial)AnyListOptionalUnion)Tensor) inception)Inception_V3_WeightsInceptionOutputs)ImageClassification)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) _fuse_modules _replace_reluquantize_model)QuantizableInception3Inception_V3_QuantizedWeights inception_v3csTeZdZdededdffdd Zdedefdd Zd d eeddfd d Z Z S)QuantizableBasicConv2dargskwargsreturnNcs tj|i|t|_dSN)super__init__nnZReLUreluselfrr __class__x/var/www/html/eduruby.in/lip-sync/lip-sync-env/lib/python3.10/site-packages/torchvision/models/quantization/inception.pyr!szQuantizableBasicConv2d.__init__xcCs"||}||}||}|SrconvZbnr#)r%r*r(r(r)forwards   zQuantizableBasicConv2d.forwardis_qatcCst|gd|dddS)Nr+T)Zinplace)r)r%r.r(r(r) fuse_model%sz!QuantizableBasicConv2d.fuse_modelr) __name__ __module__ __qualname__rr!rr-rboolr/ __classcell__r(r(r&r)rs rc<eZdZdededdffdd Zdedefdd ZZS) QuantizableInceptionArrrNc&tj|dti|tj|_dSNZ conv_blockr r!rr" quantizedFloatFunctionalmyopr$r&r(r)r!+zQuantizableInceptionA.__init__r*cC||}|j|dSNr_forwardr<catr%r*outputsr(r(r)r-/ zQuantizableInceptionA.forwardr0r1r2rr!rr-r4r(r(r&r)r6)r6cr5) QuantizableInceptionBrrrNcr7r8r9r$r&r(r)r!6r=zQuantizableInceptionB.__init__r*cCr>r?r@rCr(r(r)r-:rEzQuantizableInceptionB.forwardrFr(r(r&r)rH4rGrHcr5) QuantizableInceptionCrrrNcr7r8r9r$r&r(r)r!Ar=zQuantizableInceptionC.__init__r*cCr>r?r@rCr(r(r)r-ErEzQuantizableInceptionC.forwardrFr(r(r&r)rI?rGrIcr5) QuantizableInceptionDrrrNcr7r8r9r$r&r(r)r!Lr=zQuantizableInceptionD.__init__r*cCr>r?r@rCr(r(r)r-PrEzQuantizableInceptionD.forwardrFr(r(r&r)rJJrGrJcsReZdZdededdffdd Zdedeefdd Zdedefd d ZZ S) QuantizableInceptionErrrNcs>tj|dti|tj|_tj|_tj|_dSr8) r r!rr"r:r;myop1myop2myop3r$r&r(r)r!Ws  zQuantizableInceptionE.__init__r*cCs||}||}||||g}|j|d}||}||}||| |g}|j |d}t j |dddd}| |}||||g}|S)Nrr )Z kernel_sizeZstridepadding) branch1x1Z branch3x3_1Z branch3x3_2aZ branch3x3_2brLrBZbranch3x3dbl_1Zbranch3x3dbl_2Zbranch3x3dbl_3aZbranch3x3dbl_3brMFZ avg_pool2d branch_pool)r%r*rPZ branch3x3Z branch3x3dblrRrDr(r(r)rA]s      zQuantizableInceptionE._forwardcCr>r?)rArNrBrCr(r(r)r-rrEzQuantizableInceptionE.forward) r0r1r2rr!rrrAr-r4r(r(r&r)rKUsrKcs*eZdZdededdffdd ZZS)QuantizableInceptionAuxrrrNcstj|dti|dSr8)r r!rr$r&r(r)r!ysz QuantizableInceptionAux.__init__)r0r1r2rr!r4r(r(r&r)rSws"rScsTeZdZdededdffdd Zdedefdd Zd d ee ddfd d Z Z S)rrrrNc sDtj|dtttttttgi|t j j |_ t j j |_dS)NZinception_blocks)r r!rr6rHrIrJrKrStorchZaoZ quantizationZ QuantStubquantZ DeQuantStubdequantr$r&r(r)r!~s   zQuantizableInception3.__init__r*cCsf||}||}||\}}||}|jo|j}tjr-|s(t dt ||S| ||S)NzIScripted QuantizableInception3 always returns QuantizableInception3 Tuple) Z_transform_inputrUrArVZtraining aux_logitsrTZjitZ is_scriptingwarningswarnr Z eager_outputs)r%r*ZauxZ aux_definedr(r(r)r-s        zQuantizableInception3.forwardr.cCs(|D] }t|tur||qdS)aFuse conv/bn/relu modules in inception model Fuse conv+bn+relu/ conv+relu/conv+bn modules to prepare for quantization. Model is modified in place. Note that this operation does not change numerics and the model after modification is in floating point N)modulestyperr/)r%r.mr(r(r)r/s   z QuantizableInception3.fuse_modelr) r0r1r2rr!rr r-rr3r/r4r(r(r&r)r}s rc@sJeZdZedeedddddeddejd d d d id ddd dZ e Z dS)rzUhttps://download.pytorch.org/models/quantized/inception_v3_google_fbgemm-a2837893.pthi+iV)Z crop_sizeZ resize_sizeir)Kr]fbgemmzdhttps://github.com/pytorch/vision/tree/main/references/classification#post-training-quantized-modelsz ImageNet-1Kg%CKS@g-VW@)zacc@1zacc@5g'1@gL7A`%7@z These weights were produced by doing Post Training Quantization (eager mode) on top of the unquantized weights listed below. ) Z num_paramsZmin_size categoriesbackendZrecipeZ unquantizedZ_metricsZ_ops _file_sizeZ_docs)urlZ transformsmetaN) r0r1r2rrr rr IMAGENET1K_V1IMAGENET1K_FBGEMM_V1DEFAULTr(r(r(r)rs* rZquantized_inception_v3)nameZ pretrainedcCs|ddr tjStjS)NquantizeF)getrrer rd)rr(r(r)s rj)weightsTF)rkprogressrhrkrlrhrrcKs|rtnt|}|dd}|dur`__. .. note:: **Important**: In contrast to the other models the inception_v3 expects tensors with a size of N x 3 x 299 x 299, so ensure your images are sized accordingly. .. 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.Inception_V3_QuantizedWeights` or :class:`~torchvision.models.Inception_V3_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.Inception_V3_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, return a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableInception3`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.Inception_V3_QuantizedWeights :members: .. autoclass:: torchvision.models.Inception_V3_Weights :members: :noindex: rWFNZtransform_inputTZ num_classesr_r`r^)rlZ check_hashr()rr verifyrirlenrcpoprrrrWZ AuxLogitsZload_state_dictZget_state_dict)rkrlrhrZoriginal_aux_logitsr`modelr(r(r)rs.2      r)8rX functoolsrtypingrrrrrTZtorch.nnr"Ztorch.nn.functionalZ functionalrQrZtorchvision.modelsrZinception_moduleZtorchvision.models.inceptionr r Ztransforms._presetsr Z_apirrr_metar_utilsrrutilsrrr__all__Z BasicConv2drZ InceptionAr6Z InceptionBrHZ InceptionCrIZ InceptionDrJZ InceptionErKZ InceptionAuxrSZ Inception3rrr3rr(r(r(r)sX          ",