a  d|4@sddlZddlZddlZddlmZddlmZddlmZddlm Z m Z m Z m Z m Z mZGdddeZeGd d d ejZdS) N) ARCH_REGISTRY)nn) ResUpBlock) ConvLayer EqualConv2d EqualLinearResBlockScaledLeakyReLUStyleGAN2GeneratorBilinearcs,eZdZdZdfdd Zdd d ZZS)StyleGAN2GeneratorBilinearSFTa8StyleGAN2 Generator with SFT modulation (Spatial Feature Transform). It is the bilinear version. It does not use the complicated UpFirDnSmooth function that is not friendly for deployment. It can be easily converted to the clean version: StyleGAN2GeneratorCSFT. 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. 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. {Gz?rFcs&tt|j||||||d||_dS)N)num_style_featnum_mlpchannel_multiplierlr_mlpnarrow)superr __init__sft_half)selfout_sizerrrrrr __class__j/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/gfpgan/archs/gfpgan_bilinear_arch.pyrs z&StyleGAN2GeneratorBilinearSFT.__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) aForward function for StyleGAN2GeneratorBilinearSFT. 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).0srrr Fz9StyleGAN2GeneratorBilinearSFT.forward..Ncsg|]}tjd|qSnoise)getattrZnoises)rir!rrr"Lr#rrrr$)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'Zconv1Zconv2Znoise1Znoise2Zto_rgbZout_sameZout_sftimagerr!rforward-sV   *" z%StyleGAN2GeneratorBilinearSFT.forward)r rrrrF)FNTrNNF__name__ __module__ __qualname____doc__rr? __classcell__rrrrr s r c s,eZdZdZd fd d Zdd d ZZS)GFPGANBilinearaThe GFPGAN architecture: Unet + StyleGAN2 decoder with SFT. It is the bilinear version and it does not use the complicated UpFirDnSmooth function that is not friendly for deployment. It can be easily converted to the clean version: GFPGANv1Clean. 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. 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 rNTrrFc sRtt|||_| |_||_| 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|||}qft |_td |j d D].}|jt| d |d d d dd ddq| rtt|d d d |}n|}t| ddd|d dd dd|_t|||||| | d|_|rb|jtj|dddd|r|jD]\}}d|_qrt |_ 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 @ ) 48Z16Z32Z64Z128Z256Z512Z1024rr(rT)biasactivaterLr)ZstridepaddingrN bias_init_val)rNrRZlr_mulZ activation)rrrrrrrcSs|S)Nr)Zstoragelocrrrr#z)GFPGANBilinear.__init__..)Z map_locationZ params_emaFg?)$rrFrr8 different_wrr5mathloglog_sizerconv_body_firstrZ ModuleListconv_body_downr)r*r final_conv conv_body_uprtoRGBrr final_linearr stylegan_decoderZload_state_dictr0loadZnamed_parametersZ requires_gradcondition_scalecondition_shiftZ Sequentialr )rrrrZdecoder_load_pathZ fix_decoderrrr8rVrrZ unet_narrowZchannelsZfirst_out_sizeZ in_channelsr'Z out_channelsZlinear_out_channel_paramZsft_out_channelsrrrrs         ,    zGFPGANBilinear.__init__cCs$g}g}g}||}t|jdD]} |j| |}|d|q$||}|||dd} |j r| | dd|j } t|jdD]n} ||| }|j | |}|j | |} | | |j| |} | | |r| |j| |q|j| g|||j|d\} }| |fS)alForward function for GFPGANBilinear. 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. rrrP)r:r8r9)rZr)rYr[insertr\r_viewr6rVrr]rbr*clonercr^r`r8)rxr:Z return_rgbr9r7Z unet_skipsZout_rgbsZfeatr'Z style_codeZscaleshiftr>rdrrrr?s6     zGFPGANBilinear.forward) r rNTrrFFrF)FTTr@rrrrrFslrF)rWr.r0Zbasicsr.utils.registryrrZ gfpganv1_archrZstylegan2_bilinear_archrrrr r r r registerModulerFrrrrs    w