jz>ddlZddlZddlmcmZddlmZmZddlmZ dded e d e d e d e d e defdZ dded e d e d e d e d e defdZ ejdGddejZejdGddeZdS)N)nnTensor)_log_api_usage_onceFư>Tinputp block_sizeinplaceepstrainingreturnctjs2tjst t |dks|dkrt d|d|jdkrt d|jd|r|dkr|S|\}}}} t|| |}||z| z|dz||z d z| |z d zzzz } tj ||||z d z| |z d zf|j |j } | | tj| |dzgdzd } tj| d ||f|dz} d | z } | || zz } |r)|| | n|| z| z}|S)a Implements DropBlock2d from `"DropBlock: A regularization method for convolutional networks" `. Args: input (Tensor[N, C, H, W]): The input tensor or 4-dimensions with the first one being its batch i.e. a batch with ``N`` rows. p (float): Probability of an element to be dropped. block_size (int): Size of the block to drop. inplace (bool): If set to ``True``, will do this operation in-place. Default: ``False``. eps (float): A value added to the denominator for numerical stability. Default: 1e-6. training (bool): apply dropblock if is ``True``. Default: ``True``. Returns: Tensor[N, C, H, W]: The randomly zeroed tensor after dropblock. ?4drop probability has to be between 0 and 1, but got .z#input should be 4 dimensional. Got dimensions.rdtypedevicervalue)rrstride kernel_sizepadding)torchjit is_scripting is_tracingr drop_block2d ValueErrorndimsizeminemptyrr bernoulli_Fpad max_pool2dnumelsummul_) rr r r r r NCHWgammanoisenormalize_scales KC:\Projects\Historical\venv\Lib\site-packages\torchvision/ops/drop_block.pyr$r$ s& 9 ! ! # #*EI,@,@,B,B*L)))3ww!c''TPQTTTUUU zQWuzWWWXXX qCxx JAq!QZA&&J UQYJMq:~/Aa*nWXFX.YZ [E KAq:~11z>A3EFekbgbn o o oE U E%*/*Q.a 8 8 8E LvJ ;S]gkl]l m m mE IEkkmmsUYY[['89O0 5//// / Lctjs2tjst t |dks|dkrt d|d|jdkrt d|jd|r|dkr|S|\}}}} } t||| | }||z| z| z|dz||z d z| |z d zz| |z d zzzz } tj ||||z d z| |z d z| |z d zf|j |j } | | tj| |d zgd zd } tj| d|||f|d z} d | z } | || zz } |r)|| | n|| z| z}|S)a  Implements DropBlock3d from `"DropBlock: A regularization method for convolutional networks" `. Args: input (Tensor[N, C, D, H, W]): The input tensor or 5-dimensions with the first one being its batch i.e. a batch with ``N`` rows. p (float): Probability of an element to be dropped. block_size (int): Size of the block to drop. inplace (bool): If set to ``True``, will do this operation in-place. Default: ``False``. eps (float): A value added to the denominator for numerical stability. Default: 1e-6. training (bool): apply dropblock if is ``True``. Default: ``True``. Returns: Tensor[N, C, D, H, W]: The randomly zeroed tensor after dropblock. rrrrz#input should be 5 dimensional. Got rrrrrr)rrrr)r r!r"r#r drop_block3dr%r&r'r(r)rrr*r+r, max_pool3dr.r/r0)rr r r r r r1r2Dr3r4r5r6r7s r8r>r>7s0& 9 ! ! # #*EI,@,@,B,B*L)))3ww!c''TPQTTTUUU zQWuzWWWXXX qCxx JJLLMAq!QZAq))J UQY] A 1z>A3E!j.[\J\2]abeoaorsas2tu vE K Aq:~!1z>A#5q:~7IJRWR]fkfr   E U E%*/*Q.a 8 8 8E L ij*j-Q[eij[j   E IEkkmmsUYY[['89O0 5//// / Lr9r$c VeZdZdZddededededd f fd Zd edefd Z de fd Z xZ S) DropBlock2dz# See :func:`drop_block2d`. Frr r r r rNct||_||_||_||_dSN)super__init__r r r r selfr r r r __class__s r8rFzDropBlock2d.__init__qs9 $ r9rc\t||j|j|j|j|jSz Args: input (Tensor): Input feature map on which some areas will be randomly dropped. Returns: Tensor: The tensor after DropBlock layer. )r$r r r r r rHrs r8forwardzDropBlock2d.forwardy(E464?DL$(TXTabbbr9cT|jjd|jd|jd|jd}|S)Nz(p=z , block_size=z , inplace=))rI__name__r r r )rHss r8__repr__zDropBlock2d.__repr__s:~& j j46 j j j j[_[g j j jr9Fr) rQ __module__ __qualname____doc__floatintboolrFrrMstrrS __classcell__rIs@r8rBrBls%S4e`dcVccccc#r9rBr>c JeZdZdZd dededededd f fd Zd edefd Z xZ S) DropBlock3dz# See :func:`drop_block3d`. Frr r r r rNcPt||||dSrD)rErFrGs r8rFzDropBlock3d.__init__s' J55555r9rc\t||j|j|j|j|jSrK)r>r r r r r rLs r8rMzDropBlock3d.forwardrNr9rT) rQrUrVrWrXrYrZrFrrMr\r]s@r8r_r_s66%6S646e6`d666666cVcccccccccr9r_)FrT)r torch.fxtorch.nn.functionalr functionalr+rutilsrrXrYrZr$r>fxwrapModulerBr_r9r8rjs ''''''ko++ ++),+7;+JO+cg+ ++++^ko// //),/7;/JO/cg/ ////d n")8 nccccc+cccccr9