o p i\@svddlZddlZddlmZmZddlmZddlmZddl m Z m Z ddgZ    d d dZGd dde ZdS) N) _legacy_C_opsbase) LayerHelper) ParamAttr)Layer initializerresnet_basic_blockResNetBasicBlockFTc"@CsHtjrgd|d|d|d|d|d|d|d|d |d |d |d |d |d|d|d| d| ddd|!R}"tj|||||||||| | | | | ||||| | ||g|"R\}#}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$}$|#Stdid|d|d|d|d|d|d|d|d |d |d |d |d |d|d|d| d| d|!d(}"id)|#d*|(d+|)d,|*d-|+d.|,d/|-d0|/d1|0d2|1d3|2d4|.d5|3d6|4d7|5d8|6d9|7|8|9|:|;|<|=d:}?|%j d|>|?|"d;|#S)=Nstride1stride2stride3padding1padding2padding3 dilation1 dilation2 dilation3groupmomentumepsilon data_format has_shortcutuse_global_statstrainable_statisticsis_testact_typerelufind_conv_input_maxrT)dtype stop_gradientXZFilter1ZScale1ZBias1ZMean1ZVar1ZFilter2ZScale2ZBias2ZMean2ZVar2ZFilter3ZScale3ZBias3ZMean3ZVar3)rrYZConv1Z SavedMean1Z SavedInvstd1ZMean1OutZVar1OutZConv2Z SavedMean2Z SavedInvstd2ZMean2OutZVar2OutZ Conv2InputZConv3Z SavedMean3Z SavedInvstd3ZMean3OutZVar3Out)Z MaxInput1Z MaxFilter1Z MaxInput2Z MaxFilter2Z MaxInput3Z MaxFilter3)typeinputsoutputsattrs)r)rZ frameworkZin_dygraph_moderrrlocalscoreVarDescVarTypeFP32Z"create_variable_for_type_inferencerZ append_op)@xZfilter1Zscale1Zbias1Zmean1Zvar1Zfilter2Zscale2Zbias2Zmean2Zvar2Zfilter3Zscale3Zbias3Zmean3Zvar3r r r r rrrrrgroupsrepsrrrtrainingr find_conv_maxr%out_helperbn_param_dtypeZ max_dtypeZconv1Z saved_mean1Z saved_invstd1Z running_mean1Z running_var1Zconv2Z conv2_inputZ saved_mean2Z saved_invstd2Z running_mean2Z running_var2Zconv3Z saved_mean3Z saved_invstd3Z running_mean3Z running_var3Zconv1_input_maxZconv1_filter_maxZconv2_input_maxZconv2_filter_maxZconv3_input_maxZconv3_filter_maxr#r$r4g/home/app/PaddleOCR-VL/.venv_paddleocr/lib/python3.10/site-packages/paddle/incubate/xpu/resnet_block.pyrs $      !"#$%&A                      csleZdZdZ                              dfd d Zd dZZS)r a ResNetBasicBlock is designed for optimize the performance of the basic unit of ssd resnet block. If has_shortcut = True, it can calculate 3 Conv2D, 3 BatchNorm and 2 ReLU in one time. If has_shortcut = False, it can calculate 2 Conv2D, 2 BatchNorm and 2 ReLU in one time. In this case the shape of output is same with input. Args: num_channels (int): The number of input image channel. num_filter (int): The number of filter. It is as same as the output image channel. filter_size (int|list|tuple): The filter size. If filter_size is a tuple, it must contain two integers, (filter_size_height, filter_size_width). Otherwise, filter_size_height = filter_size_width =\ filter_size. stride (int, optional): The stride size. It means the stride in convolution. If stride is a tuple, it must contain two integers, (stride_height, stride_width). Otherwise, stride_height = stride_width = stride. Default: stride = 1. act (str, optional): Activation type, if it is set to None, activation is not appended. Default: None momentum (float, optional): The value used for the moving_mean and moving_var computation. This should be a float number or a 0-D Tensor with shape [] and data type as float32. The updated formula is: :math:`moving\_mean = moving\_mean * momentum + new\_mean * (1. - momentum)` :math:`moving\_var = moving\_var * momentum + new\_var * (1. - momentum)` Default is 0.9. eps (float, optional): A value added to the denominator for numerical stability. Default is 1e-5. data_format (str, optional): Specify the data format of the input, and the data format of the output will be consistent with that of the input. Now is only support `"NCHW"`, the data is stored in the order of: `[batch_size, input_channels, input_height, input_width]`. has_shortcut (bool, optional): Whether to calculate CONV3 and BN3. Default: False. use_global_stats (bool, optional): Whether to use global mean and variance. In inference or test mode, set use_global_stats to true or is_test to true, and the behavior is equivalent. In train mode, when setting use_global_stats True, the global mean and variance are also used during train period. Default: False. is_test (bool, optional): A flag indicating whether it is in test phrase or not. Default: False. filter_attr (ParamAttr, optional): The parameter attribute for learnable parameters/weights of conv2d. If it is set to None or one attribute of ParamAttr, conv2d will create ParamAttr as param_attr. Default: None. scale_attr (ParamAttr, optional): The parameter attribute for Parameter `scale` of batch_norm. If it is set to None or one attribute of ParamAttr, batch_norm will create ParamAttr as param_attr, the name of scale can be set in ParamAttr. If the Initializer of the param_attr is not set, the parameter is initialized with Xavier. Default: None. bias_attr (ParamAttr, optional): The parameter attribute for the bias of batch_norm. If it is set to None or one attribute of ParamAttr, batch_norm will create ParamAttr as bias_attr, the name of bias can be set in ParamAttr. If the Initializer of the bias_attr is not set, the bias is initialized zero. Default: None. moving_mean_name (str, optional): The name of moving_mean which store the global Mean. If it is set to None, batch_norm will save global mean with a random name, otherwise, batch_norm will save global mean with the string. Default: None. moving_var_name (str, optional): The name of the moving_variance which store the global Variance. If it is set to None, batch_norm will save global variance with a random name, otherwise, batch_norm will save global variance with the string. Default: None. padding (int, optional): The padding size. It is only support padding_height = padding_width = padding. Default: padding = 0. dilation (int, optional): The dilation size. It means the spacing between the kernel points. It is only support dilation_height = dilation_width = dilation. Default: dilation = 1. trainable_statistics (bool, optional): Whether to calculate mean and var in eval mode. In eval mode, when setting trainable_statistics True, mean and variance will be calculated by current batch statistics. Default: False. find_conv_max (bool, optional): Whether to calculate max value of each conv2d. Default: True. Returns: A Tensor representing the ResNetBasicBlock, whose data type is the same with input. Examples: .. code-block:: python >>> # doctest: +REQUIRES(env: XPU) >>> import paddle >>> from paddle.incubate.xpu.resnet_block import ResNetBasicBlock >>> ch_in = 4 >>> ch_out = 8 >>> x = paddle.uniform((2, ch_in, 16, 16), dtype='float32', min=-1., max=1.) >>> resnet_basic_block = ResNetBasicBlock(num_channels1=ch_in, ... num_filter1=ch_out, ... filter1_size=3, ... num_channels2=ch_out, ... num_filter2=ch_out, ... filter2_size=3, ... num_channels3=ch_in, ... num_filter3=ch_out, ... filter3_size=1, ... stride1=1, ... stride2=1, ... stride3=1, ... act='relu', ... padding1=1, ... padding2=1, ... padding3=0, ... has_shortcut=True) >>> out = resnet_basic_block.forward(x) >>> print(out.shape) [2, 8, 16, 16] r?h㈵>NCHWFNrTc+4sHt| |_| |_tj|dd|_tj|dd|_|&|_ |'|_ |#|_ |$|_ d|_ ||_||_||_| |_||_||_||_|)|_|*|_|r\tj| dd|_|%|_| |_|(|_n d|_d|_d|_d|_dh}+||+vrytd|+d|d d },tjjjj }-dd|g}.dd|g}/||||g}0||||g}1|j!|0||,||jd |_"|j!|.||-t#$d d |_%|j!|.||-dd|_&|j!t'|t#$ddd|.|-d|_(d|j(_)|j!t'|t#$d dd|.|-d|_*d|j*_)|j!|1||,||jd |_+|j!|/||-t#$d d |_,|j!|/||-dd|_-|j!t'|t#$ddd|/|-d|_.d|j._)|j!t'|t#$d dd|/|-d|_/d|j/_)|rdd|g}2||| | g}3|j!|3||,||jd |_0|j!|2||-t#$d d |_1|j!|2| |-dd|_2|j!t'|!t#$ddd|2|-d|_3d|j3_)|j!t'|"t#$d dd|2|-d|_4d|j4_)dSd|_0d|_1d|_2d|_3d|_4dS)N filter1_size filter2_sizer6 filter3_sizer9zconv_format must be one of z, but got conv_format=cSs&t||}d|d}td|S)Ng@g?)npprodIZNormal)ZchannelsZ kernel_sizeZfilter_elem_numZstdr4r4r5_get_default_param_initializer s  zAResNetBasicBlock.__init__.._get_default_param_initializer)shapeattrdefault_initializerg?)rCrDrrET)rCrDrZis_biasr>F)namerZ trainable)rDrCr)5super__init___stride1_stride2paddleutilsZconvert_to_listZ _kernel1_sizeZ _kernel2_size _dilation1 _dilation2 _padding1 _padding2_groups _momentum_eps _data_formatZ_act _has_shortcut_use_global_statsZ_is_test_trainable_statistics_find_conv_maxZ _kernel3_size _padding3_stride3 _dilation3 ValueErrorrr'r(r)r*Zcreate_parameterfilter_1rAConstantscale_1bias_1rmean_1rvar_1filter_2scale_2bias_2mean_2var_2filter_3scale_3bias_3mean_3var_3)4selfZ num_channels1Z num_filter1r;Z num_channels2Z num_filter2r<Z num_channels3Z num_filter3r=r r r Zactrr-rrrrZ filter1_attrZ scale1_attrZ bias1_attrZmoving_mean1_nameZmoving_var1_nameZ filter2_attrZ scale2_attrZ bias2_attrZmoving_mean2_nameZmoving_var2_nameZ filter3_attrZ scale3_attrZ bias3_attrZmoving_mean3_nameZmoving_var3_namer rrrrrrr/Z valid_formatrBr3Zbn1_param_shapeZbn2_param_shapeZ filter1_shapeZ filter2_shapeZbn3_param_shapeZ filter3_shape __class__r4r5rHs\ -             zResNetBasicBlock.__init__cCst||j|j|j|j|j|j|j|j|j |j |j |j |j |j|j|j|j|j|j|j|j|j|j|j|j|j|j|j|jf|j|j|j |j!d}|S)N)rr.rr/)"rr]r_r`rarbrcrdrerfrgrhrirjrkrlrIrJrZrOrPrYrMrNr[rQrRrSrTrUrVr.rWrX)rmr+r0r4r4r5forwardsL$zResNetBasicBlock.forward)!r6r6r6rr7r8r9FFFNNNNNNNNNNNNNNNrrrr6r6r6FT)__name__ __module__ __qualname____doc__rHrp __classcell__r4r4rnr5r GsLul)NFFT)numpyr?rKrrZpaddle.base.layer_helperrZpaddle.base.param_attrrZ paddle.nnrrrA__all__rr r4r4r4r5s  " ,