a  d @sddlZddlmZddlmZddlmZmZGdddejZ Gdddej Z Gd d d ej Z Gd d d ej Z Gd ddej ZeGdddej ZdS)N) ARCH_REGISTRY)ResidualBlockNoBN make_layercs"eZdZdZdfdd ZZS) MeanShifta Data normalization with mean and std. Args: rgb_range (int): Maximum value of RGB. rgb_mean (list[float]): Mean for RGB channels. rgb_std (list[float]): Std for RGB channels. sign (int): For subtraction, sign is -1, for addition, sign is 1. Default: -1. requires_grad (bool): Whether to update the self.weight and self.bias. Default: True. Tcstt|jddddt|}tddddd|j_|jj |dddd||t||j _|j j |||_ dS)Nr) kernel_size) superr__init__torchZTensorZeyeviewZweightdataZdiv_Zbias requires_grad)selfZ rgb_rangergb_meanrgb_stdsignrZstd __class__b/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/basicsr/archs/ridnet_arch.pyr s zMeanShift.__init__)rT)__name__ __module__ __qualname____doc__r __classcell__rrrrrs rcs(eZdZdZfddZddZZS)EResidualBlockNoBNzEnhanced Residual block without BN. There are three convolution layers in residual branch. It has a style of: ---Conv-ReLU-Conv-ReLU-Conv-+-ReLU- |__________________________| c sntt|tt||dddtjddt||dddtjddt||ddd|_tjdd|_dS)NrrTZinplacer) r rr nn SequentialConv2dReLUbodyrelu)r in_channels out_channelsrrrr )s  zEResidualBlockNoBN.__init__cCs||}|||}|SN)r#r$rxoutrrrforward5s zEResidualBlockNoBN.forwardrrrrr r+rrrrrrs rcs*eZdZdZdfdd ZddZZS) MergeRuna Merge-and-run unit. This unit contains two branches with different dilated convolutions, followed by a convolution to process the concatenated features. Paper: Real Image Denoising with Feature Attention Ref git repo: https://github.com/saeed-anwar/RIDNet rrc stt|tt|||||tjddt||||ddtjdd|_tt||||ddtjddt||||ddtjdd|_tt|d||||tjdd|_ dS)NTrr) r r-r rr r!r" dilation1 dilation2 aggregation)rr%r&r Zstridepaddingrrrr EszMergeRun.__init__cCs<||}||}tj||gdd}||}||}|S)Nr)Zdim)r0r1r catr2)rr)r0r1r*rrrr+Rs    zMergeRun.forward)rrrr,rrrrr-;s  r-cs*eZdZdZdfdd ZddZZS)ChannelAttentionzChannel attention. Args: num_feat (int): Channel number of intermediate features. squeeze_factor (int): Channel squeeze factor. Default: c s\tt|ttdtj|||dddtjddtj|||dddt|_ dS)Nrr)r3Tr) r r5r rr ZAdaptiveAvgPool2dr!r"ZSigmoid attention)r mid_channelsZsqueeze_factorrrrr cs $zChannelAttention.__init__cCs||}||Sr')r7)rr)yrrrr+is zChannelAttention.forward)r6r,rrrrr5[sr5cs(eZdZdZfddZddZZS)EAMakEnhancement attention modules (EAM) in RIDNet. This module contains a merge-and-run unit, a residual block, an enhanced residual block and a feature attention unit. Attributes: merge: The merge-and-run unit. block1: The residual block. block2: The enhanced residual block. ca: The feature/channel attention unit. csLtt|t|||_t||_t|||_t ||_ t j dd|_ dS)NTr)r r:r r-mergerblock1rblock2r5carr"r$)rr%r8r&rrrr {s     z EAM.__init__cCs2||}|||}||}||}|Sr')r;r$r<r=r>r(rrrr+s    z EAM.forwardr,rrrrr:ns r:cs*eZdZdZd fdd Zdd ZZS) RIDNeta0RIDNet: Real Image Denoising with Feature Attention. Ref git repo: https://github.com/saeed-anwar/RIDNet Args: in_channels (int): Channel number of inputs. mid_channels (int): Channel number of EAM modules. Default: 64. out_channels (int): Channel number of outputs. num_block (int): Number of EAM. Default: 4. img_range (float): Image range. Default: 255. rgb_mean (tuple[float]): Image mean in RGB orders. Default: (0.4488, 0.4371, 0.4040), calculated from DIV2K dataset. r/o@gw#?g8EGr?gB`"??rCrCcsztt|t||||_t|||d|_t||ddd|_t t ||||d|_ t||ddd|_ tj dd|_dS)Nrr)r%r8r&Tr)r r?r rsub_meanadd_meanrr!headrr:r#tailr"r$)rr%r8r&Z num_blockZ img_rangerrrrrr s zRIDNet.__init__c Cs<||}|||||}||}||}|Sr')rDrGr#r$rFrE)rr)resr*rrrr+s   zRIDNet.forward)r/r@rArBr,rrrrr?sr?)r Ztorch.nnrZbasicsr.utils.registryrZ arch_utilrrr!rModulerr-r5r:registerr?rrrrs