3j{SSKJr SSKrSSKJs Jr SSKJr "SS\R5r"SS\R5r "SS \R5r "S S \R5r SS jr SS jrg)) annotationsN)nncf^\rSrSrSrSSU4SjjjrSrU=r$) ConvNormActaApply a sequence of Convolution, Normalization, and Activation layers. Args: in_channels: Number of input channels. out_channels: Number of output channels. kernel_size: Size of the convolution kernel. stride: Stride of the convolution. Default: 1. padding: Zero-padding added to both sides of the input. Default: 0. groups: Number of blocked connections from input to output. Default: 1. norm: Normalization layer type. Default: :class:`nn.BatchNorm2d`. act: Activation layer type. Default: :class:`nn.ReLU`. c >[TU]5 US-S:Xa0US- S-n US- U - n [R"XX45UlSn OUS- S-n [R "XX4U SUS5n [R "U5n[R[R[RS.U"SS9nURX}5 URX5 URX5 g)NrF)relusilunoneT)inplace) super__init__r ZeroPad2dpadConv2d BatchNorm2dReLUSiLUIdentity __setattr__)self in_channels out_channels kernel_sizestrideactgroups conv_naming norm_naming act_namingp1p2paddingconvnorm activation __class__s N/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/models/common.pyrConvNormAct.__init__'s  ?a  /a'Bq2%B||RR$45DHG"Q1,GyyKRSU[]bc~~l+ ggrww LSQZ^_  + + 0)r)r r r r&r'r)rintrr-rr-rr-rstrrr-r r.r!r.r"r.returnNone)__name__ __module__ __qualname____firstlineno____doc__r__static_attributes__ __classcell__r)s@r*rrs $!!111 1  1  11111 11r,rcV^\rSrSrSrSSU4SjjjrSSjrSrU=r$) MLPJa]Implement a Multi-Layer Perceptron (MLP) for feature projection. Args: input_dim: The number of input features. hidden_dim: The number of features in hidden layers. output_dim: The number of output features. num_layers: The total number of layers. sigmoid_output: Whether to apply a sigmoid to the final output. Default: False. c>[TU]5 X@lU/US- -n[R"S[ U/UQ/UQUP555UlXPlg)Nr c3R# UHup[R"X5v M g7fN)rLinear).0nks r* MLP.__init__..[s#g@fBIIaOO@fs%')rr num_layersr ModuleListziplayerssigmoid_output)r input_dim hidden_dim output_dimrErIhr)s r*r MLP.__init__Us\ $ LJN +mm#gYOQROUeWXUeZdUe@f#gg ,r,c[UR5H;up#X RS- :a[R"U"U55OU"U5nM= UR (a[R "U5nU$)Nr ) enumeraterHrEFr rIsigmoid)rxilayers r*forward MLP.forward^sX!$++.HA$%!(;$;uQx qA/    ! Ar,)rHrErI)F) rJr-rKr-rLr-rEr-rIboolr/r0rS torch.Tensorr/rZ r1r2r3r4r5rrVr6r7r8s@r*r:r:JsSin--*--;>-LO-ae- --r,r:c>^\rSrSrSrSSU4SjjjrSSjrSrU=r$) DropPathhzYDrop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).c:>[TU]5 XlX lgr>)rr drop_prob scale_by_keep)rr`rar)s r*rDropPath.__init__ks "*r,cDURS:XdUR(dU$SUR- nURS4SURS- --nUR U5R U5nUS:a"UR (aURU5 X-$)Nr r)r )r`trainingshapendim new_empty bernoulli_radiv_)rrS keep_probrf random_tensors r*rVDropPath.forwardps >>S  H&   33 E*55i@ s?t11   y )  r,)r`ra)rdT)r`floatrarXr/r0rYr[r8s@r*r]r]hsc++ !!r,r]c>^\rSrSrSrSSU4SjjjrSSjrSrU=r$) LayerNorm2d}z8Apply Layer Normalization over a 4D input tensor (NCHW).c>[TU]5 [R"[R "U55Ul[R"[R"U55UlX l gr>) rrr Parametertorchonesweightzerosbiaseps)r num_channelsryr)s r*rLayerNorm2d.__init__sE ll5::l#;< LL\!:; r,cURSSS9nX- RS5RSSS9nX- X0R-R5- nURSS2SS4U-UR SS2SS4-nU$)Nr T)keepdimr )meanpowrysqrtrvrx)rrSuss r*rVLayerNorm2d.forwards FF1dF # UKKN  4  0 Uq88|))+ + KK4 & *TYYq$}-E Er,)rxryrv)gư>)rzr-ryrnr/r0rYr[r8s@r*rprp}sB r,rpc NURup#pEXU-- U-nXU-- U-nUS:dUS:a[R"USSSUSU45nX6-XG-pURX(U-XU-X5nUR SSSSSS5R 5RSXU5n XU 44$)a'Partition into non-overlapping windows with padding if needed. Args: x: input tokens with [B, H, W, C]. window_size: window size. Returns: windows: windows after partition with [B * num_windows, window_size, window_size, C]. (Hp, Wp): padded height and width before partition rr r )rfrQrviewpermute contiguous) rS window_sizeBHWCpad_hpad_wHpWpwindowss r*window_partitionrsJA! {? *k 9E {? *k 9E qyEAI EE!aAua/ 0 Y  q #[ 2C[TAii1aAq)446;;B Z[\G H r,c8UupEUupgURSXE-U-U--nURXU-XQ-XS5n U RSSSSSS5R5RXUS5n XF:dXW:aU SS2SU2SU2SS24R5n U $) aYWindow unpartition into original sequences and removing padding. Args: windows: input tokens with [B * num_windows, window_size, window_size, C]. window_size: window size. pad_hw: padded height and width (Hp, Wp). hw: original height and width (H, W) before padding. Returns: x: unpartitioned sequences with [B, H, W, C]. rrr rr rrN)rfrrr) rrpad_hwhwrrrrrrSs r*window_unpartitionrsFB DA aRW 3{BCA Qk)2+rs$# ,1"--,1b"))<!ryy!*"))"2  (+ 5D JY  r,