a  du<@sdZddlmZddlZddlmZddlmZmZddl m Z m Z m Z m Z mZmZmZddlmZddlmZdd lmZmZd gZGd d d ejZGd ddejZGdddejZGdd d ejZeegdgdgddgdddddegdgdgddgdddddegdgdgddgdd d d!degdgdgddgdd d d!degdgdgddgdd dd!degdgdgddgdd dd!degdgdgddgdd dd!degdgdgddgdd dd!degdgdgddgd"d dd!degdgdgddgdd dd#dd$ Z e d%e d&<dDd'd(Z!dEd)d*Z"ee"dd+e"dd+e"dd+e"d,d-d.d/e"dd+e"d,d-d.d/e"dd+e"dd+e"dd+e"dd+d0 Z#edFed1d2d3Z$edGed1d4d5Z%edHed1d6d7Z&edIed1d8d9Z'edJed1d:d;Z(edKed1dZ*edMed1d?d@Z+edNed1dAdBZ,edOed1dCd&Z-dS)Pa VoVNet (V1 & V2) Papers: * `An Energy and GPU-Computation Efficient Backbone Network` - https://arxiv.org/abs/1904.09730 * `CenterMask : Real-Time Anchor-Free Instance Segmentation` - https://arxiv.org/abs/1911.06667 Looked at https://github.com/youngwanLEE/vovnet-detectron2 & https://github.com/stigma0617/VoVNet.pytorch/blob/master/models_vovnet/vovnet.py for some reference, rewrote most of the code. Hacked together by / Copyright 2020 Ross Wightman )ListNIMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD) ConvNormActSeparableConvNormActBatchNormAct2dClassifierHeadDropPath create_attncreate_norm_act_layer)build_model_with_cfg)checkpoint_seq)register_modelgenerate_default_cfgsVovNetcs8eZdZfddZejeejejdddZZS)SequentialAppendListcstt|j|dSN)superr__init__)selfargs __class__[/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/timm/models/vovnet.pyrszSequentialAppendList.__init__)x concat_listreturncCsNt|D]2\}}|dkr(|||q|||dqtj|dd}|S)Nrr )Zdim) enumerateappendtorchcat)rrrimodulerrrforward!s zSequentialAppendList.forward) __name__ __module__ __qualname__rr#ZTensorrr' __classcell__rrrrrs rcs4eZdZdddeejdffdd ZddZZS)OsaBlockFNc stt|||_||_t|| d} |} |jrX| |krX|r@Jt| |dfi| |_nd|_g} t|D]@}|jrt ||fi| }nt| |dfi| }|} | |qjt | |_ |||} t| |fi| |_ |rt||nd|_| |_dS)N norm_layer act_layerr )rr,rresidual depthwisedictrconv_reductionrangerr"rconv_mid conv_concatr attn drop_path)rin_chsmid_chsout_chslayer_per_blockr2r3r9r/r0r: conv_kwargsZ next_in_chsZ mid_convsr%convrrrr-s*      zOsaBlock.__init__cCsn|g}|jdur||}|||}||}|jdurD||}|jdurX||}|jrj||d}|S)Nr)r5r7r8r9r:r2)rroutputrrrr'Ys         zOsaBlock.forward r(r)r*rnnReLUrr'r+rrrrr,+s,r,cs6eZdZddddeejdffdd ZddZZS) OsaStageTFeseNc stt|d|_|r,tjdddd|_nd|_g} t|D]l}||dk}| durp| |dkrpt| |}nd}| t |||||o|dk||r| nd | | |d g7} |}q>tj | |_ dS) NFr1T)Z kernel_sizestrideZ ceil_moder rr-)r2r3r9r/r0r:) rrErgrad_checkpointingrCZ MaxPool2dpoolr6r r, Sequentialblocks)rr;r<r=block_per_stager> downsampler2r3r9r/r0drop_path_ratesrMr%Z last_blockr:rrrrjs"   zOsaStage.__init__cCs@|jdur||}|jr2tjs2t|j|}n ||}|Sr)rKrJr#jitZ is_scriptingrrMrrrrrr's    zOsaStage.forwardrBrrrrrEhs %rEcseZdZddddeejddffdd Zejj dd d Z ejj dd d Z ejj ddZ dddZ ddZdedddZddZZS)rr1avg rIc  stt|||_||_|dks&Jt|fi| }|dd} |d} |d} |d}|d}|d}t||d }| d }|d rtnt}t j t|| d d fdd i||| d | dd fddi||| d| d d fd|i|g|_ t| dd d| dkr dnd dg|_ | }t t d | t||}| dd|dd}tf|d|d |dd|}g}tdD]}| d kp|d k}|t||| ||||||f|||d|g7}|||_||rd nd9}|j t|j|d|dg7_ q~t j ||_t|j|||d|_|D]J\}}t|t jrbt jj|jdddnt|t jr6t j|jq6dS)a Args: cfg (dict): Model architecture configuration in_chans (int): Number of input channels (default: 3) num_classes (int): Number of classifier classes (default: 1000) global_pool (str): Global pooling type (default: 'avg') output_stride (int): Output stride of network, one of (8, 16, 32) (default: 32) norm_layer (Union[str, nn.Module]): normalization layer act_layer (Union[str, nn.Module]): activation layer drop_rate (float): Dropout rate (default: 0.) drop_path_rate (float): Stochastic depth drop-path rate (default: 0.) kwargs (dict): Extra kwargs overlayed onto cfg rU stem_stridestem_chsstage_conv_chs stage_out_chsrNr>r.rGr3rr1rHr zstem.)Znum_chsZ reductionr&r Nr2r9)r2r3r9)rOrPzstages.Z pool_type drop_rateZfan_outZrelu)modeZ nonlinearity) rrr num_classesr\r4getrrrCrLstemZ feature_infor#splitZlinspacesumr6rE num_featuresstagesr headZ named_modules isinstanceZConv2dinitZkaiming_normal_ZweightZLinearZzeros_Zbias)rcfgZin_chansr^ global_poolZ output_strider/r0r\Zdrop_path_ratekwargsrVrXrYrZrNr>r?Zlast_stem_strideZ conv_typeZcurrent_strideZ stage_dprZ in_ch_listZ stage_argsrdr%rOnmrrrrsf    "  & zVovNet.__init__FcCstd|r dnddS)Nz^stemz^stages\.(\d+)z^stages\.(\d+).blocks\.(\d+))r`rM)r4)rZcoarserrr group_matchers zVovNet.group_matcherTcCs|jD] }||_qdSr)rdrJ)renablesrrrset_grad_checkpointings zVovNet.set_grad_checkpointingcCs|jjSr)reZfc)rrrrget_classifierszVovNet.get_classifiercCst|j|||jd|_dS)Nr[)r rcr\re)rr^rirrrreset_classifierszVovNet.reset_classifiercCs||}||Sr)r`rdrRrrrforward_featuress zVovNet.forward_features pre_logitscCs|j||dS)Nrt)re)rrrurrr forward_headszVovNet.forward_headcCs||}||}|Sr)rsrvrRrrrr's  zVovNet.forward)F)T)rT)F)r(r)r*rrCrDrr#rQignorermrprqrrrsboolrvr'r+rrrrrs&S    )@ry)rz)ii)r r rGrGFr-)rXrYrZr>rNr2r3r9)r r rWr1)ryryry)ryP`p)rr~irr1)r r r r TrF)r r1 r1Zeca) vovnet39a vovnet57aese_vovnet19b_slim_dwese_vovnet19b_dwese_vovnet19b_slimZ ese_vovnet19b ese_vovnet39b ese_vovnet57b ese_vovnet99b eca_vovnet39brese_vovnet39b_evoscKs$tt||ft|tddd|S)NT)Zflatten_sequential)Z model_cfgZ feature_cfg)rr model_cfgsr4)variant pretrainedrjrrr_create_vovnet}src Ks|dddddttddd |S) NrS)r1r}r})rg?Zbicubicz stem.0.convzhead.fc) urlr^Z input_sizeZ pool_sizeZcrop_pct interpolationmeanZstdZ first_conv classifierr)rrjrrr_cfgsr)rztimm/)r1 rgffffff?)Z hf_hub_idZtest_input_sizeZ test_crop_pct) zvovnet39a.untrainedzvovnet57a.untrainedzese_vovnet19b_slim_dw.untrainedzese_vovnet19b_dw.ra_in1kzese_vovnet19b_slim.untrainedzese_vovnet39b.ra_in1kzese_vovnet57b.untrainedzese_vovnet99b.untrainedzeca_vovnet39b.untrainedzese_vovnet39b_evos.untrained)rcKstdd|i|S)Nrr)rrrrjrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrscKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKstdd|i|S)Nrr)rrrrrrrsrcKsdd}td||d|S)Nc[std|fddi|S)NZ evonorms0rQF)r )rcnkwargsrrr norm_act_fnsz'ese_vovnet39b_evos..norm_act_fnr)rr/)rr)rrjrrrrrs)F)r-)F)F)F)F)F)F)F)F)F)F).__doc__typingrr#Ztorch.nnrCZ timm.datarrZ timm.layersrrrr r r r Z_builderrZ _manipulater _registryrr__all__rLrModuler,rErr4rrrZ default_cfgsrrrrrrrrrrrrrrs8  $  =1w         j