a  dG@sddlZddlZddlZddlmZmZmZmZmZm Z ddl m Z ddl m Z ddlmZddlmZGddde ZGd d d ejZGd d d ejZe Gd ddejZe GdddejZdS)N) ConvLayer EqualConv2d EqualLinearResBlockScaledLeakyReLUStyleGAN2Generator)FusedLeakyReLU) ARCH_REGISTRY)nn) functionalcs,eZdZdZdfd d Zdd dZZS)StyleGAN2GeneratorSFTaAStyleGAN2 Generator with SFT modulation (Spatial Feature Transform). Args: out_size (int): The spatial size of outputs. num_style_feat (int): Channel number of style features. Default: 512. num_mlp (int): Layer number of MLP style layers. Default: 8. channel_multiplier (int): Channel multiplier for large networks of StyleGAN2. Default: 2. resample_kernel (list[int]): A list indicating the 1D resample kernel magnitude. A cross production will be applied to extent 1D resample kernel to 2D resample kernel. Default: (1, 3, 3, 1). lr_mlp (float): Learning rate multiplier for mlp layers. Default: 0.01. narrow (float): The narrow ratio for channels. Default: 1. sft_half (bool): Whether to apply SFT on half of the input channels. Default: False. rr{Gz?rFc s(tt|j|||||||d||_dS)N)num_style_featnum_mlpchannel_multiplierresample_kernellr_mlpnarrow)superr __init__sft_half) selfout_sizerrrrrrr __class__c/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/gfpgan/archs/gfpganv1_arch.pyrs zStyleGAN2GeneratorSFT.__init__NTc s|sfdd|D}|durH|r0dgj}nfddtjD}|dkrxg} |D]} | ||| |qX| }t|dkrj}|djdkr|ddd|d} n|d} nnt|dkr*|durt djd}|ddd|d} |dddj|d}t | |gd} | j d}j|| dddf|dd }|| dddf}d}tjdddjddd|ddd|dddjD]\}}}}}||| dd|f|d }|t|krXjr@t j|t|dddd \}}|||d||}t j ||gdd }n|||d||}||| dd|df|d }||| dd|df|}|d7}q|}| r|| fS|dfSdS) a Forward function for StyleGAN2GeneratorSFT. Args: styles (list[Tensor]): Sample codes of styles. conditions (list[Tensor]): SFT conditions to generators. input_is_latent (bool): Whether input is latent style. Default: False. noise (Tensor | None): Input noise or None. Default: None. randomize_noise (bool): Randomize noise, used when 'noise' is False. Default: True. truncation (float): The truncation ratio. Default: 1. truncation_latent (Tensor | None): The truncation latent tensor. Default: None. inject_index (int | None): The injection index for mixing noise. Default: None. return_latents (bool): Whether to return style latents. Default: False. csg|]}|qSr!)Z style_mlp).0srr!r" Gz1StyleGAN2GeneratorSFT.forward..Ncsg|]}tjd|qSnoise)getattrZnoises)r#ir%r!r"r&Mr'rrrrr()Zdim)Z num_layersrangeappendlenZ num_latentndimZ unsqueezerepeatrandomrandinttorchcatZconstant_inputshapeZ style_conv1Zto_rgb1zipZ style_convsZto_rgbsrsplitintsize)rZstyles conditionsinput_is_latentr)randomize_noiseZ truncationZtruncation_latentZ inject_indexreturn_latentsZstyle_truncationstyleZlatentZlatent1Zlatent2outskipr+conv1conv2Znoise1Znoise2Zto_rgbZout_sameZout_sftimager!r%r"forward.sV   *" zStyleGAN2GeneratorSFT.forward)r rrrrrF)FNTrNNF__name__ __module__ __qualname____doc__rrD __classcell__r!r!rr"r s"r cs*eZdZdZd fdd ZddZZS) ConvUpLayerajConvolutional upsampling layer. It uses bilinear upsampler + Conv. Args: in_channels (int): Channel number of the input. out_channels (int): Channel number of the output. kernel_size (int): Size of the convolving kernel. stride (int): Stride of the convolution. Default: 1 padding (int): Zero-padding added to both sides of the input. Default: 0. bias (bool): If ``True``, adds a learnable bias to the output. Default: ``True``. bias_init_val (float): Bias initialized value. Default: 0. activate (bool): Whether use activateion. Default: True. rrTc stt|||_||_||_||_||_dt ||d|_ t t |||||_|r~|s~t t |||_n |dd|r|rt||_qtd|_nd|_dS)Nrrbias皙?)rrKr in_channels out_channels kernel_sizestridepaddingmathsqrtscaler Parameterr3ZrandnweightzerosZfill_rLZregister_parameterr activationr) rrNrOrPrQrRrL bias_init_valactivaterr!r"rs    zConvUpLayer.__init__cCsLtj|dddd}tj||j|j|j|j|jd}|jdurH||}|S)NrZbilinearF)Z scale_factormodeZ align_corners)rLrQrR) FZ interpolateZconv2drWrUrLrQrRrY)rxr?r!r!r"rDs   zConvUpLayer.forward)rrTrTrEr!r!rr"rKs"rKcs(eZdZdZfddZddZZS) ResUpBlockzResidual block with upsampling. Args: in_channels (int): Channel number of the input. out_channels (int): Channel number of the output. c sRtt|t||dddd|_t||dddddd|_t||dddd|_dS)NrTrLr[r)rQrRrLr[F)rr_rrrArKrBr@)rrNrOrr!r"rszResUpBlock.__init__cCs4||}||}||}||td}|S)Nr)rArBr@rSrT)rr^r?r@r!r!r"rDs    zResUpBlock.forwardrEr!r!rr"r_s r_c s,eZdZdZdfd d Zdd d ZZS)GFPGANv1aThe GFPGAN architecture: Unet + StyleGAN2 decoder with SFT. Ref: GFP-GAN: Towards Real-World Blind Face Restoration with Generative Facial Prior. Args: out_size (int): The spatial size of outputs. num_style_feat (int): Channel number of style features. Default: 512. channel_multiplier (int): Channel multiplier for large networks of StyleGAN2. Default: 2. resample_kernel (list[int]): A list indicating the 1D resample kernel magnitude. A cross production will be applied to extent 1D resample kernel to 2D resample kernel. Default: (1, 3, 3, 1). decoder_load_path (str): The path to the pre-trained decoder model (usually, the StyleGAN2). Default: None. fix_decoder (bool): Whether to fix the decoder. Default: True. num_mlp (int): Layer number of MLP style layers. Default: 8. lr_mlp (float): Learning rate multiplier for mlp layers. Default: 0.01. input_is_latent (bool): Whether input is latent style. Default: False. different_w (bool): Whether to use different latent w for different layers. Default: False. narrow (float): The narrow ratio for channels. Default: 1. sft_half (bool): Whether to apply SFT on half of the input channels. Default: False. r rrNTrrFc sVtt|| |_| |_||_| d} td| td| td| td| td|| td|| td|| td|| td|| d }tt|d |_ d tt|d }t d ||d d d d |_ ||}t |_t|j d dD].}|d |d }|jt||||}qt ||dd d d d |_|d}t |_td |j d D]*}|d |}|jt|||}qht |_td |j d D].}|jt|d |d d d dd ddq| rtt|d d d |}n|}t|ddd|d dd dd|_t||||||| | d|_|rf|jtj|dddd|r|jD]\}}d|_qvt |_ t |_!td |j d D]}|d |}| r|}n|d }|j t "t||d d d d ddt#dt||d d d d d d|j!t "t||d d d d ddt#dt||d d d d ddqdS)Ng?r @ ) 48Z16Z32Z64Z128Z256Z512Z1024rrrTr`rgr)rQrRrLrZ)rLrZZlr_mulrY)rrrrrrrrcSs|S)Nr!)Zstoragelocr!r!r"Gr'z#GFPGANv1.__init__..)Z map_locationZ params_emaFrM)$rrarr; different_wrr8rSloglog_sizerconv_body_firstr Z ModuleListconv_body_downr,r-r final_conv conv_body_upr_toRGBrr final_linearr stylegan_decoderZload_state_dictr3loadZnamed_parametersZ requires_gradcondition_scalecondition_shiftZ Sequentialr)rrrrrZdecoder_load_pathZ fix_decoderrrr;rmrrZ unet_narrowZchannelsZfirst_out_sizerNr+rOZlinear_out_channel_paramZsft_out_channelsrr!r"rs         ,    zGFPGANv1.__init__cKs$g}g}g}||} t|jdD]} |j| | } |d| q$|| } || | dd} |j r| | dd|j } t|jdD]n} | || } |j | | } |j | | } | | |j| | } | | |r| |j| | q|j| g|||j|d\}}||fS)afForward function for GFPGANv1. Args: x (Tensor): Input images. return_latents (bool): Whether to return style latents. Default: False. return_rgb (bool): Whether return intermediate rgb images. Default: True. randomize_noise (bool): Randomize noise, used when 'noise' is False. Default: True. rrri)r=r;r<)rpr,rorqinsertrrruviewr9rmrrsrxr-cloneryrtrvr;)rr^r=Z return_rgbr<kwargsr:Z unet_skipsZout_rgbsfeatr+Z style_coderUshiftrCrzr!r!r"rDas6     zGFPGANv1.forward) r rrNTrrFFrF)FTTrEr!r!rr"rasnracs*eZdZdZfddZdddZZS)FacialComponentDiscriminatorzHFacial component (eyes, mouth, noise) discriminator used in GFPGAN. c stt|tdddddddd|_tdddddddd|_tdddddddd|_tdddddddd|_tdddddddd|_tdd dddd |_ dS) NrrdFrT)Z downsamplerrLr[rcrbrr`) rrrrrArBconv3conv4conv5rrr%rr!r"rsz%FacialComponentDiscriminator.__init__FcKst||}|||}g}|r0|||||}|rR||||}|rh||fS|dfSdS)zForward function for FacialComponentDiscriminator. Args: x (Tensor): Input images. return_feats (bool): Whether to return intermediate features. Default: False. N)rArrBr-r~rrrr)rr^Z return_featsrrZ rlt_featsr?r!r!r"rDs  z$FacialComponentDiscriminator.forward)FrEr!r!rr"rs r)rSr1r3Zbasicsr.archs.stylegan2_archrrrrrrZbasicsr.ops.fused_actrZbasicsr.utils.registryr r Ztorch.nnr r]r ModulerKr_registerrarr!r!r!r"s     xA8