Skip to content

Instantly share code, notes, and snippets.

@ayush1999

ayush1999/densenet121.jl

Created July 14, 2018 09:22
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save ayush1999/079f73c5b0b654b9e5d629389cfe89f3 to your computer and use it in GitHub Desktop.
Save ayush1999/079f73c5b0b654b9e5d629389cfe89f3 to your computer and use it in GitHub Desktop.
Download ZIP
DenseNet121
Raw
densenet121.jl
flipkernel(x) = x[end:-1:1, end:-1:1, :, :]
begin
edge_1 = (Flux.unsqueeze)(weights["conv1/bn_w_0"], (ndims(weights["conv1/bn_w_0"]) + 1) - 1)
edge_2 = (Flux.unsqueeze)(weights["conv1/bn_b_0"], (ndims(weights["conv1/bn_b_0"]) + 1) - 1)
edge_3 = (Flux.unsqueeze)(weights["conv2_1/x1/bn_w_0"], (ndims(weights["conv2_1/x1/bn_w_0"]) + 1) - 1)
edge_4 = (Flux.unsqueeze)(weights["conv2_1/x1/bn_b_0"], (ndims(weights["conv2_1/x1/bn_b_0"]) + 1) - 1)
edge_5 = (Flux.unsqueeze)(weights["conv2_1/x2/bn_w_0"], (ndims(weights["conv2_1/x2/bn_w_0"]) + 1) - 1)
edge_6 = (Flux.unsqueeze)(weights["conv2_1/x2/bn_b_0"], (ndims(weights["conv2_1/x2/bn_b_0"]) + 1) - 1)
edge_7 = (Flux.unsqueeze)(weights["conv2_2/x1/bn_w_0"], (ndims(weights["conv2_2/x1/bn_w_0"]) + 1) - 1)
edge_8 = (Flux.unsqueeze)(weights["conv2_2/x1/bn_b_0"], (ndims(weights["conv2_2/x1/bn_b_0"]) + 1) - 1)
edge_9 = (Flux.unsqueeze)(weights["conv2_2/x2/bn_w_0"], (ndims(weights["conv2_2/x2/bn_w_0"]) + 1) - 1)
edge_10 = (Flux.unsqueeze)(weights["conv2_2/x2/bn_b_0"], (ndims(weights["conv2_2/x2/bn_b_0"]) + 1) - 1)
edge_11 = (Flux.unsqueeze)(weights["conv2_3/x1/bn_w_0"], (ndims(weights["conv2_3/x1/bn_w_0"]) + 1) - 1)
edge_12 = (Flux.unsqueeze)(weights["conv2_3/x1/bn_b_0"], (ndims(weights["conv2_3/x1/bn_b_0"]) + 1) - 1)
edge_13 = (Flux.unsqueeze)(weights["conv2_3/x2/bn_w_0"], (ndims(weights["conv2_3/x2/bn_w_0"]) + 1) - 1)
edge_14 = (Flux.unsqueeze)(weights["conv2_3/x2/bn_b_0"], (ndims(weights["conv2_3/x2/bn_b_0"]) + 1) - 1)
edge_15 = (Flux.unsqueeze)(weights["conv2_4/x1/bn_w_0"], (ndims(weights["conv2_4/x1/bn_w_0"]) + 1) - 1)
edge_16 = (Flux.unsqueeze)(weights["conv2_4/x1/bn_b_0"], (ndims(weights["conv2_4/x1/bn_b_0"]) + 1) - 1)
edge_17 = (Flux.unsqueeze)(weights["conv2_4/x2/bn_w_0"], (ndims(weights["conv2_4/x2/bn_w_0"]) + 1) - 1)
edge_18 = (Flux.unsqueeze)(weights["conv2_4/x2/bn_b_0"], (ndims(weights["conv2_4/x2/bn_b_0"]) + 1) - 1)
edge_19 = (Flux.unsqueeze)(weights["conv2_5/x1/bn_w_0"], (ndims(weights["conv2_5/x1/bn_w_0"]) + 1) - 1)
edge_20 = (Flux.unsqueeze)(weights["conv2_5/x1/bn_b_0"], (ndims(weights["conv2_5/x1/bn_b_0"]) + 1) - 1)
edge_21 = (Flux.unsqueeze)(weights["conv2_5/x2/bn_w_0"], (ndims(weights["conv2_5/x2/bn_w_0"]) + 1) - 1)
edge_22 = (Flux.unsqueeze)(weights["conv2_5/x2/bn_b_0"], (ndims(weights["conv2_5/x2/bn_b_0"]) + 1) - 1)
edge_23 = (Flux.unsqueeze)(weights["conv2_6/x1/bn_w_0"], (ndims(weights["conv2_6/x1/bn_w_0"]) + 1) - 1)
edge_24 = (Flux.unsqueeze)(weights["conv2_6/x1/bn_b_0"], (ndims(weights["conv2_6/x1/bn_b_0"]) + 1) - 1)
edge_25 = (Flux.unsqueeze)(weights["conv2_6/x2/bn_w_0"], (ndims(weights["conv2_6/x2/bn_w_0"]) + 1) - 1)
edge_26 = (Flux.unsqueeze)(weights["conv2_6/x2/bn_b_0"], (ndims(weights["conv2_6/x2/bn_b_0"]) + 1) - 1)
edge_27 = (Flux.unsqueeze)(weights["conv2_blk/bn_w_0"], (ndims(weights["conv2_blk/bn_w_0"]) + 1) - 1)
edge_28 = (Flux.unsqueeze)(weights["conv2_blk/bn_b_0"], (ndims(weights["conv2_blk/bn_b_0"]) + 1) - 1)
edge_29 = (Flux.unsqueeze)(weights["conv3_1/x1/bn_w_0"], (ndims(weights["conv3_1/x1/bn_w_0"]) + 1) - 1)
edge_30 = (Flux.unsqueeze)(weights["conv3_1/x1/bn_b_0"], (ndims(weights["conv3_1/x1/bn_b_0"]) + 1) - 1)
edge_31 = (Flux.unsqueeze)(weights["conv3_1/x2/bn_w_0"], (ndims(weights["conv3_1/x2/bn_w_0"]) + 1) - 1)
edge_32 = (Flux.unsqueeze)(weights["conv3_1/x2/bn_b_0"], (ndims(weights["conv3_1/x2/bn_b_0"]) + 1) - 1)
edge_33 = (Flux.unsqueeze)(weights["conv3_2/x1/bn_w_0"], (ndims(weights["conv3_2/x1/bn_w_0"]) + 1) - 1)
edge_34 = (Flux.unsqueeze)(weights["conv3_2/x1/bn_b_0"], (ndims(weights["conv3_2/x1/bn_b_0"]) + 1) - 1)
edge_35 = (Flux.unsqueeze)(weights["conv3_2/x2/bn_w_0"], (ndims(weights["conv3_2/x2/bn_w_0"]) + 1) - 1)
edge_36 = (Flux.unsqueeze)(weights["conv3_2/x2/bn_b_0"], (ndims(weights["conv3_2/x2/bn_b_0"]) + 1) - 1)
edge_37 = (Flux.unsqueeze)(weights["conv3_3/x1/bn_w_0"], (ndims(weights["conv3_3/x1/bn_w_0"]) + 1) - 1)
edge_38 = (Flux.unsqueeze)(weights["conv3_3/x1/bn_b_0"], (ndims(weights["conv3_3/x1/bn_b_0"]) + 1) - 1)
edge_39 = (Flux.unsqueeze)(weights["conv3_3/x2/bn_w_0"], (ndims(weights["conv3_3/x2/bn_w_0"]) + 1) - 1)
edge_40 = (Flux.unsqueeze)(weights["conv3_3/x2/bn_b_0"], (ndims(weights["conv3_3/x2/bn_b_0"]) + 1) - 1)
edge_41 = (Flux.unsqueeze)(weights["conv3_4/x1/bn_w_0"], (ndims(weights["conv3_4/x1/bn_w_0"]) + 1) - 1)
edge_42 = (Flux.unsqueeze)(weights["conv3_4/x1/bn_b_0"], (ndims(weights["conv3_4/x1/bn_b_0"]) + 1) - 1)
edge_43 = (Flux.unsqueeze)(weights["conv3_4/x2/bn_w_0"], (ndims(weights["conv3_4/x2/bn_w_0"]) + 1) - 1)
edge_44 = (Flux.unsqueeze)(weights["conv3_4/x2/bn_b_0"], (ndims(weights["conv3_4/x2/bn_b_0"]) + 1) - 1)
edge_45 = (Flux.unsqueeze)(weights["conv3_5/x1/bn_w_0"], (ndims(weights["conv3_5/x1/bn_w_0"]) + 1) - 1)
edge_46 = (Flux.unsqueeze)(weights["conv3_5/x1/bn_b_0"], (ndims(weights["conv3_5/x1/bn_b_0"]) + 1) - 1)
edge_47 = (Flux.unsqueeze)(weights["conv3_5/x2/bn_w_0"], (ndims(weights["conv3_5/x2/bn_w_0"]) + 1) - 1)
edge_48 = (Flux.unsqueeze)(weights["conv3_5/x2/bn_b_0"], (ndims(weights["conv3_5/x2/bn_b_0"]) + 1) - 1)
edge_49 = (Flux.unsqueeze)(weights["conv3_6/x1/bn_w_0"], (ndims(weights["conv3_6/x1/bn_w_0"]) + 1) - 1)
edge_50 = (Flux.unsqueeze)(weights["conv3_6/x1/bn_b_0"], (ndims(weights["conv3_6/x1/bn_b_0"]) + 1) - 1)
edge_51 = (Flux.unsqueeze)(weights["conv3_6/x2/bn_w_0"], (ndims(weights["conv3_6/x2/bn_w_0"]) + 1) - 1)
edge_52 = (Flux.unsqueeze)(weights["conv3_6/x2/bn_b_0"], (ndims(weights["conv3_6/x2/bn_b_0"]) + 1) - 1)
edge_53 = (Flux.unsqueeze)(weights["conv3_7/x1/bn_w_0"], (ndims(weights["conv3_7/x1/bn_w_0"]) + 1) - 1)
edge_54 = (Flux.unsqueeze)(weights["conv3_7/x1/bn_b_0"], (ndims(weights["conv3_7/x1/bn_b_0"]) + 1) - 1)
edge_55 = (Flux.unsqueeze)(weights["conv3_7/x2/bn_w_0"], (ndims(weights["conv3_7/x2/bn_w_0"]) + 1) - 1)
edge_56 = (Flux.unsqueeze)(weights["conv3_7/x2/bn_b_0"], (ndims(weights["conv3_7/x2/bn_b_0"]) + 1) - 1)
edge_57 = (Flux.unsqueeze)(weights["conv3_8/x1/bn_w_0"], (ndims(weights["conv3_8/x1/bn_w_0"]) + 1) - 1)
edge_58 = (Flux.unsqueeze)(weights["conv3_8/x1/bn_b_0"], (ndims(weights["conv3_8/x1/bn_b_0"]) + 1) - 1)
edge_59 = (Flux.unsqueeze)(weights["conv3_8/x2/bn_w_0"], (ndims(weights["conv3_8/x2/bn_w_0"]) + 1) - 1)
edge_60 = (Flux.unsqueeze)(weights["conv3_8/x2/bn_b_0"], (ndims(weights["conv3_8/x2/bn_b_0"]) + 1) - 1)
edge_61 = (Flux.unsqueeze)(weights["conv3_9/x1/bn_w_0"], (ndims(weights["conv3_9/x1/bn_w_0"]) + 1) - 1)
edge_62 = (Flux.unsqueeze)(weights["conv3_9/x1/bn_b_0"], (ndims(weights["conv3_9/x1/bn_b_0"]) + 1) - 1)
edge_63 = (Flux.unsqueeze)(weights["conv3_9/x2/bn_w_0"], (ndims(weights["conv3_9/x2/bn_w_0"]) + 1) - 1)
edge_64 = (Flux.unsqueeze)(weights["conv3_9/x2/bn_b_0"], (ndims(weights["conv3_9/x2/bn_b_0"]) + 1) - 1)
edge_65 = (Flux.unsqueeze)(weights["conv3_10/x1/bn_w_0"], (ndims(weights["conv3_10/x1/bn_w_0"]) + 1) - 1)
edge_66 = (Flux.unsqueeze)(weights["conv3_10/x1/bn_b_0"], (ndims(weights["conv3_10/x1/bn_b_0"]) + 1) - 1)
edge_67 = (Flux.unsqueeze)(weights["conv3_10/x2/bn_w_0"], (ndims(weights["conv3_10/x2/bn_w_0"]) + 1) - 1)
edge_68 = (Flux.unsqueeze)(weights["conv3_10/x2/bn_b_0"], (ndims(weights["conv3_10/x2/bn_b_0"]) + 1) - 1)
edge_69 = (Flux.unsqueeze)(weights["conv3_11/x1/bn_w_0"], (ndims(weights["conv3_11/x1/bn_w_0"]) + 1) - 1)
edge_70 = (Flux.unsqueeze)(weights["conv3_11/x1/bn_b_0"], (ndims(weights["conv3_11/x1/bn_b_0"]) + 1) - 1)
edge_71 = (Flux.unsqueeze)(weights["conv3_11/x2/bn_w_0"], (ndims(weights["conv3_11/x2/bn_w_0"]) + 1) - 1)
edge_72 = (Flux.unsqueeze)(weights["conv3_11/x2/bn_b_0"], (ndims(weights["conv3_11/x2/bn_b_0"]) + 1) - 1)
edge_73 = (Flux.unsqueeze)(weights["conv3_12/x1/bn_w_0"], (ndims(weights["conv3_12/x1/bn_w_0"]) + 1) - 1)
edge_74 = (Flux.unsqueeze)(weights["conv3_12/x1/bn_b_0"], (ndims(weights["conv3_12/x1/bn_b_0"]) + 1) - 1)
edge_75 = (Flux.unsqueeze)(weights["conv3_12/x2/bn_w_0"], (ndims(weights["conv3_12/x2/bn_w_0"]) + 1) - 1)
edge_76 = (Flux.unsqueeze)(weights["conv3_12/x2/bn_b_0"], (ndims(weights["conv3_12/x2/bn_b_0"]) + 1) - 1)
edge_77 = (Flux.unsqueeze)(weights["conv3_blk/bn_w_0"], (ndims(weights["conv3_blk/bn_w_0"]) + 1) - 1)
edge_78 = (Flux.unsqueeze)(weights["conv3_blk/bn_b_0"], (ndims(weights["conv3_blk/bn_b_0"]) + 1) - 1)
edge_79 = (Flux.unsqueeze)(weights["conv4_1/x1/bn_w_0"], (ndims(weights["conv4_1/x1/bn_w_0"]) + 1) - 1)
edge_80 = (Flux.unsqueeze)(weights["conv4_1/x1/bn_b_0"], (ndims(weights["conv4_1/x1/bn_b_0"]) + 1) - 1)
edge_81 = (Flux.unsqueeze)(weights["conv4_1/x2/bn_w_0"], (ndims(weights["conv4_1/x2/bn_w_0"]) + 1) - 1)
edge_82 = (Flux.unsqueeze)(weights["conv4_1/x2/bn_b_0"], (ndims(weights["conv4_1/x2/bn_b_0"]) + 1) - 1)
edge_83 = (Flux.unsqueeze)(weights["conv4_2/x1/bn_w_0"], (ndims(weights["conv4_2/x1/bn_w_0"]) + 1) - 1)
edge_84 = (Flux.unsqueeze)(weights["conv4_2/x1/bn_b_0"], (ndims(weights["conv4_2/x1/bn_b_0"]) + 1) - 1)
edge_85 = (Flux.unsqueeze)(weights["conv4_2/x2/bn_w_0"], (ndims(weights["conv4_2/x2/bn_w_0"]) + 1) - 1)
edge_86 = (Flux.unsqueeze)(weights["conv4_2/x2/bn_b_0"], (ndims(weights["conv4_2/x2/bn_b_0"]) + 1) - 1)
edge_87 = (Flux.unsqueeze)(weights["conv4_3/x1/bn_w_0"], (ndims(weights["conv4_3/x1/bn_w_0"]) + 1) - 1)
edge_88 = (Flux.unsqueeze)(weights["conv4_3/x1/bn_b_0"], (ndims(weights["conv4_3/x1/bn_b_0"]) + 1) - 1)
edge_89 = (Flux.unsqueeze)(weights["conv4_3/x2/bn_w_0"], (ndims(weights["conv4_3/x2/bn_w_0"]) + 1) - 1)
edge_90 = (Flux.unsqueeze)(weights["conv4_3/x2/bn_b_0"], (ndims(weights["conv4_3/x2/bn_b_0"]) + 1) - 1)
edge_91 = (Flux.unsqueeze)(weights["conv4_4/x1/bn_w_0"], (ndims(weights["conv4_4/x1/bn_w_0"]) + 1) - 1)
edge_92 = (Flux.unsqueeze)(weights["conv4_4/x1/bn_b_0"], (ndims(weights["conv4_4/x1/bn_b_0"]) + 1) - 1)
edge_93 = (Flux.unsqueeze)(weights["conv4_4/x2/bn_w_0"], (ndims(weights["conv4_4/x2/bn_w_0"]) + 1) - 1)
edge_94 = (Flux.unsqueeze)(weights["conv4_4/x2/bn_b_0"], (ndims(weights["conv4_4/x2/bn_b_0"]) + 1) - 1)
edge_95 = (Flux.unsqueeze)(weights["conv4_5/x1/bn_w_0"], (ndims(weights["conv4_5/x1/bn_w_0"]) + 1) - 1)
edge_96 = (Flux.unsqueeze)(weights["conv4_5/x1/bn_b_0"], (ndims(weights["conv4_5/x1/bn_b_0"]) + 1) - 1)
edge_97 = (Flux.unsqueeze)(weights["conv4_5/x2/bn_w_0"], (ndims(weights["conv4_5/x2/bn_w_0"]) + 1) - 1)
edge_98 = (Flux.unsqueeze)(weights["conv4_5/x2/bn_b_0"], (ndims(weights["conv4_5/x2/bn_b_0"]) + 1) - 1)
edge_99 = (Flux.unsqueeze)(weights["conv4_6/x1/bn_w_0"], (ndims(weights["conv4_6/x1/bn_w_0"]) + 1) - 1)
edge_100 = (Flux.unsqueeze)(weights["conv4_6/x1/bn_b_0"], (ndims(weights["conv4_6/x1/bn_b_0"]) + 1) - 1)
edge_101 = (Flux.unsqueeze)(weights["conv4_6/x2/bn_w_0"], (ndims(weights["conv4_6/x2/bn_w_0"]) + 1) - 1)
edge_102 = (Flux.unsqueeze)(weights["conv4_6/x2/bn_b_0"], (ndims(weights["conv4_6/x2/bn_b_0"]) + 1) - 1)
edge_103 = (Flux.unsqueeze)(weights["conv4_7/x1/bn_w_0"], (ndims(weights["conv4_7/x1/bn_w_0"]) + 1) - 1)
edge_104 = (Flux.unsqueeze)(weights["conv4_7/x1/bn_b_0"], (ndims(weights["conv4_7/x1/bn_b_0"]) + 1) - 1)
edge_105 = (Flux.unsqueeze)(weights["conv4_7/x2/bn_w_0"], (ndims(weights["conv4_7/x2/bn_w_0"]) + 1) - 1)
edge_106 = (Flux.unsqueeze)(weights["conv4_7/x2/bn_b_0"], (ndims(weights["conv4_7/x2/bn_b_0"]) + 1) - 1)
edge_107 = (Flux.unsqueeze)(weights["conv4_8/x1/bn_w_0"], (ndims(weights["conv4_8/x1/bn_w_0"]) + 1) - 1)
edge_108 = (Flux.unsqueeze)(weights["conv4_8/x1/bn_b_0"], (ndims(weights["conv4_8/x1/bn_b_0"]) + 1) - 1)
edge_109 = (Flux.unsqueeze)(weights["conv4_8/x2/bn_w_0"], (ndims(weights["conv4_8/x2/bn_w_0"]) + 1) - 1)
edge_110 = (Flux.unsqueeze)(weights["conv4_8/x2/bn_b_0"], (ndims(weights["conv4_8/x2/bn_b_0"]) + 1) - 1)
edge_111 = (Flux.unsqueeze)(weights["conv4_9/x1/bn_w_0"], (ndims(weights["conv4_9/x1/bn_w_0"]) + 1) - 1)
edge_112 = (Flux.unsqueeze)(weights["conv4_9/x1/bn_b_0"], (ndims(weights["conv4_9/x1/bn_b_0"]) + 1) - 1)
edge_113 = (Flux.unsqueeze)(weights["conv4_9/x2/bn_w_0"], (ndims(weights["conv4_9/x2/bn_w_0"]) + 1) - 1)
edge_114 = (Flux.unsqueeze)(weights["conv4_9/x2/bn_b_0"], (ndims(weights["conv4_9/x2/bn_b_0"]) + 1) - 1)
edge_115 = (Flux.unsqueeze)(weights["conv4_10/x1/bn_w_0"], (ndims(weights["conv4_10/x1/bn_w_0"]) + 1) - 1)
edge_116 = (Flux.unsqueeze)(weights["conv4_10/x1/bn_b_0"], (ndims(weights["conv4_10/x1/bn_b_0"]) + 1) - 1)
edge_117 = (Flux.unsqueeze)(weights["conv4_10/x2/bn_w_0"], (ndims(weights["conv4_10/x2/bn_w_0"]) + 1) - 1)
edge_118 = (Flux.unsqueeze)(weights["conv4_10/x2/bn_b_0"], (ndims(weights["conv4_10/x2/bn_b_0"]) + 1) - 1)
edge_119 = (Flux.unsqueeze)(weights["conv4_11/x1/bn_w_0"], (ndims(weights["conv4_11/x1/bn_w_0"]) + 1) - 1)
edge_120 = (Flux.unsqueeze)(weights["conv4_11/x1/bn_b_0"], (ndims(weights["conv4_11/x1/bn_b_0"]) + 1) - 1)
edge_121 = (Flux.unsqueeze)(weights["conv4_11/x2/bn_w_0"], (ndims(weights["conv4_11/x2/bn_w_0"]) + 1) - 1)
edge_122 = (Flux.unsqueeze)(weights["conv4_11/x2/bn_b_0"], (ndims(weights["conv4_11/x2/bn_b_0"]) + 1) - 1)
edge_123 = (Flux.unsqueeze)(weights["conv4_12/x1/bn_w_0"], (ndims(weights["conv4_12/x1/bn_w_0"]) + 1) - 1)
edge_124 = (Flux.unsqueeze)(weights["conv4_12/x1/bn_b_0"], (ndims(weights["conv4_12/x1/bn_b_0"]) + 1) - 1)
edge_125 = (Flux.unsqueeze)(weights["conv4_12/x2/bn_w_0"], (ndims(weights["conv4_12/x2/bn_w_0"]) + 1) - 1)
edge_126 = (Flux.unsqueeze)(weights["conv4_12/x2/bn_b_0"], (ndims(weights["conv4_12/x2/bn_b_0"]) + 1) - 1)
edge_127 = (Flux.unsqueeze)(weights["conv4_13/x1/bn_w_0"], (ndims(weights["conv4_13/x1/bn_w_0"]) + 1) - 1)
edge_128 = (Flux.unsqueeze)(weights["conv4_13/x1/bn_b_0"], (ndims(weights["conv4_13/x1/bn_b_0"]) + 1) - 1)
edge_129 = (Flux.unsqueeze)(weights["conv4_13/x2/bn_w_0"], (ndims(weights["conv4_13/x2/bn_w_0"]) + 1) - 1)
edge_130 = (Flux.unsqueeze)(weights["conv4_13/x2/bn_b_0"], (ndims(weights["conv4_13/x2/bn_b_0"]) + 1) - 1)
edge_131 = (Flux.unsqueeze)(weights["conv4_14/x1/bn_w_0"], (ndims(weights["conv4_14/x1/bn_w_0"]) + 1) - 1)
edge_132 = (Flux.unsqueeze)(weights["conv4_14/x1/bn_b_0"], (ndims(weights["conv4_14/x1/bn_b_0"]) + 1) - 1)
edge_133 = (Flux.unsqueeze)(weights["conv4_14/x2/bn_w_0"], (ndims(weights["conv4_14/x2/bn_w_0"]) + 1) - 1)
edge_134 = (Flux.unsqueeze)(weights["conv4_14/x2/bn_b_0"], (ndims(weights["conv4_14/x2/bn_b_0"]) + 1) - 1)
edge_135 = (Flux.unsqueeze)(weights["conv4_15/x1/bn_w_0"], (ndims(weights["conv4_15/x1/bn_w_0"]) + 1) - 1)
edge_136 = (Flux.unsqueeze)(weights["conv4_15/x1/bn_b_0"], (ndims(weights["conv4_15/x1/bn_b_0"]) + 1) - 1)
edge_137 = (Flux.unsqueeze)(weights["conv4_15/x2/bn_w_0"], (ndims(weights["conv4_15/x2/bn_w_0"]) + 1) - 1)
edge_138 = (Flux.unsqueeze)(weights["conv4_15/x2/bn_b_0"], (ndims(weights["conv4_15/x2/bn_b_0"]) + 1) - 1)
edge_139 = (Flux.unsqueeze)(weights["conv4_16/x1/bn_w_0"], (ndims(weights["conv4_16/x1/bn_w_0"]) + 1) - 1)
edge_140 = (Flux.unsqueeze)(weights["conv4_16/x1/bn_b_0"], (ndims(weights["conv4_16/x1/bn_b_0"]) + 1) - 1)
edge_141 = (Flux.unsqueeze)(weights["conv4_16/x2/bn_w_0"], (ndims(weights["conv4_16/x2/bn_w_0"]) + 1) - 1)
edge_142 = (Flux.unsqueeze)(weights["conv4_16/x2/bn_b_0"], (ndims(weights["conv4_16/x2/bn_b_0"]) + 1) - 1)
edge_143 = (Flux.unsqueeze)(weights["conv4_17/x1/bn_w_0"], (ndims(weights["conv4_17/x1/bn_w_0"]) + 1) - 1)
edge_144 = (Flux.unsqueeze)(weights["conv4_17/x1/bn_b_0"], (ndims(weights["conv4_17/x1/bn_b_0"]) + 1) - 1)
edge_145 = (Flux.unsqueeze)(weights["conv4_17/x2/bn_w_0"], (ndims(weights["conv4_17/x2/bn_w_0"]) + 1) - 1)
edge_146 = (Flux.unsqueeze)(weights["conv4_17/x2/bn_b_0"], (ndims(weights["conv4_17/x2/bn_b_0"]) + 1) - 1)
edge_147 = (Flux.unsqueeze)(weights["conv4_18/x1/bn_w_0"], (ndims(weights["conv4_18/x1/bn_w_0"]) + 1) - 1)
edge_148 = (Flux.unsqueeze)(weights["conv4_18/x1/bn_b_0"], (ndims(weights["conv4_18/x1/bn_b_0"]) + 1) - 1)
edge_149 = (Flux.unsqueeze)(weights["conv4_18/x2/bn_w_0"], (ndims(weights["conv4_18/x2/bn_w_0"]) + 1) - 1)
edge_150 = (Flux.unsqueeze)(weights["conv4_18/x2/bn_b_0"], (ndims(weights["conv4_18/x2/bn_b_0"]) + 1) - 1)
edge_151 = (Flux.unsqueeze)(weights["conv4_19/x1/bn_w_0"], (ndims(weights["conv4_19/x1/bn_w_0"]) + 1) - 1)
edge_152 = (Flux.unsqueeze)(weights["conv4_19/x1/bn_b_0"], (ndims(weights["conv4_19/x1/bn_b_0"]) + 1) - 1)
edge_153 = (Flux.unsqueeze)(weights["conv4_19/x2/bn_w_0"], (ndims(weights["conv4_19/x2/bn_w_0"]) + 1) - 1)
edge_154 = (Flux.unsqueeze)(weights["conv4_19/x2/bn_b_0"], (ndims(weights["conv4_19/x2/bn_b_0"]) + 1) - 1)
edge_155 = (Flux.unsqueeze)(weights["conv4_20/x1/bn_w_0"], (ndims(weights["conv4_20/x1/bn_w_0"]) + 1) - 1)
edge_156 = (Flux.unsqueeze)(weights["conv4_20/x1/bn_b_0"], (ndims(weights["conv4_20/x1/bn_b_0"]) + 1) - 1)
edge_157 = (Flux.unsqueeze)(weights["conv4_20/x2/bn_w_0"], (ndims(weights["conv4_20/x2/bn_w_0"]) + 1) - 1)
edge_158 = (Flux.unsqueeze)(weights["conv4_20/x2/bn_b_0"], (ndims(weights["conv4_20/x2/bn_b_0"]) + 1) - 1)
edge_159 = (Flux.unsqueeze)(weights["conv4_21/x1/bn_w_0"], (ndims(weights["conv4_21/x1/bn_w_0"]) + 1) - 1)
edge_160 = (Flux.unsqueeze)(weights["conv4_21/x1/bn_b_0"], (ndims(weights["conv4_21/x1/bn_b_0"]) + 1) - 1)
edge_161 = (Flux.unsqueeze)(weights["conv4_21/x2/bn_w_0"], (ndims(weights["conv4_21/x2/bn_w_0"]) + 1) - 1)
edge_162 = (Flux.unsqueeze)(weights["conv4_21/x2/bn_b_0"], (ndims(weights["conv4_21/x2/bn_b_0"]) + 1) - 1)
edge_163 = (Flux.unsqueeze)(weights["conv4_22/x1/bn_w_0"], (ndims(weights["conv4_22/x1/bn_w_0"]) + 1) - 1)
edge_164 = (Flux.unsqueeze)(weights["conv4_22/x1/bn_b_0"], (ndims(weights["conv4_22/x1/bn_b_0"]) + 1) - 1)
edge_165 = (Flux.unsqueeze)(weights["conv4_22/x2/bn_w_0"], (ndims(weights["conv4_22/x2/bn_w_0"]) + 1) - 1)
edge_166 = (Flux.unsqueeze)(weights["conv4_22/x2/bn_b_0"], (ndims(weights["conv4_22/x2/bn_b_0"]) + 1) - 1)
edge_167 = (Flux.unsqueeze)(weights["conv4_23/x1/bn_w_0"], (ndims(weights["conv4_23/x1/bn_w_0"]) + 1) - 1)
edge_168 = (Flux.unsqueeze)(weights["conv4_23/x1/bn_b_0"], (ndims(weights["conv4_23/x1/bn_b_0"]) + 1) - 1)
edge_169 = (Flux.unsqueeze)(weights["conv4_23/x2/bn_w_0"], (ndims(weights["conv4_23/x2/bn_w_0"]) + 1) - 1)
edge_170 = (Flux.unsqueeze)(weights["conv4_23/x2/bn_b_0"], (ndims(weights["conv4_23/x2/bn_b_0"]) + 1) - 1)
edge_171 = (Flux.unsqueeze)(weights["conv4_24/x1/bn_w_0"], (ndims(weights["conv4_24/x1/bn_w_0"]) + 1) - 1)
edge_172 = (Flux.unsqueeze)(weights["conv4_24/x1/bn_b_0"], (ndims(weights["conv4_24/x1/bn_b_0"]) + 1) - 1)
edge_173 = (Flux.unsqueeze)(weights["conv4_24/x2/bn_w_0"], (ndims(weights["conv4_24/x2/bn_w_0"]) + 1) - 1)
edge_174 = (Flux.unsqueeze)(weights["conv4_24/x2/bn_b_0"], (ndims(weights["conv4_24/x2/bn_b_0"]) + 1) - 1)
edge_175 = (Flux.unsqueeze)(weights["conv4_blk/bn_w_0"], (ndims(weights["conv4_blk/bn_w_0"]) + 1) - 1)
edge_176 = (Flux.unsqueeze)(weights["conv4_blk/bn_b_0"], (ndims(weights["conv4_blk/bn_b_0"]) + 1) - 1)
edge_177 = (Flux.unsqueeze)(weights["conv5_1/x1/bn_w_0"], (ndims(weights["conv5_1/x1/bn_w_0"]) + 1) - 1)
edge_178 = (Flux.unsqueeze)(weights["conv5_1/x1/bn_b_0"], (ndims(weights["conv5_1/x1/bn_b_0"]) + 1) - 1)
edge_179 = (Flux.unsqueeze)(weights["conv5_1/x2/bn_w_0"], (ndims(weights["conv5_1/x2/bn_w_0"]) + 1) - 1)
edge_180 = (Flux.unsqueeze)(weights["conv5_1/x2/bn_b_0"], (ndims(weights["conv5_1/x2/bn_b_0"]) + 1) - 1)
edge_181 = (Flux.unsqueeze)(weights["conv5_2/x1/bn_w_0"], (ndims(weights["conv5_2/x1/bn_w_0"]) + 1) - 1)
edge_182 = (Flux.unsqueeze)(weights["conv5_2/x1/bn_b_0"], (ndims(weights["conv5_2/x1/bn_b_0"]) + 1) - 1)
edge_183 = (Flux.unsqueeze)(weights["conv5_2/x2/bn_w_0"], (ndims(weights["conv5_2/x2/bn_w_0"]) + 1) - 1)
edge_184 = (Flux.unsqueeze)(weights["conv5_2/x2/bn_b_0"], (ndims(weights["conv5_2/x2/bn_b_0"]) + 1) - 1)
edge_185 = (Flux.unsqueeze)(weights["conv5_3/x1/bn_w_0"], (ndims(weights["conv5_3/x1/bn_w_0"]) + 1) - 1)
edge_186 = (Flux.unsqueeze)(weights["conv5_3/x1/bn_b_0"], (ndims(weights["conv5_3/x1/bn_b_0"]) + 1) - 1)
edge_187 = (Flux.unsqueeze)(weights["conv5_3/x2/bn_w_0"], (ndims(weights["conv5_3/x2/bn_w_0"]) + 1) - 1)
edge_188 = (Flux.unsqueeze)(weights["conv5_3/x2/bn_b_0"], (ndims(weights["conv5_3/x2/bn_b_0"]) + 1) - 1)
edge_189 = (Flux.unsqueeze)(weights["conv5_4/x1/bn_w_0"], (ndims(weights["conv5_4/x1/bn_w_0"]) + 1) - 1)
edge_190 = (Flux.unsqueeze)(weights["conv5_4/x1/bn_b_0"], (ndims(weights["conv5_4/x1/bn_b_0"]) + 1) - 1)
edge_191 = (Flux.unsqueeze)(weights["conv5_4/x2/bn_w_0"], (ndims(weights["conv5_4/x2/bn_w_0"]) + 1) - 1)
edge_192 = (Flux.unsqueeze)(weights["conv5_4/x2/bn_b_0"], (ndims(weights["conv5_4/x2/bn_b_0"]) + 1) - 1)
edge_193 = (Flux.unsqueeze)(weights["conv5_5/x1/bn_w_0"], (ndims(weights["conv5_5/x1/bn_w_0"]) + 1) - 1)
edge_194 = (Flux.unsqueeze)(weights["conv5_5/x1/bn_b_0"], (ndims(weights["conv5_5/x1/bn_b_0"]) + 1) - 1)
edge_195 = (Flux.unsqueeze)(weights["conv5_5/x2/bn_w_0"], (ndims(weights["conv5_5/x2/bn_w_0"]) + 1) - 1)
edge_196 = (Flux.unsqueeze)(weights["conv5_5/x2/bn_b_0"], (ndims(weights["conv5_5/x2/bn_b_0"]) + 1) - 1)
edge_197 = (Flux.unsqueeze)(weights["conv5_6/x1/bn_w_0"], (ndims(weights["conv5_6/x1/bn_w_0"]) + 1) - 1)
edge_198 = (Flux.unsqueeze)(weights["conv5_6/x1/bn_b_0"], (ndims(weights["conv5_6/x1/bn_b_0"]) + 1) - 1)
edge_199 = (Flux.unsqueeze)(weights["conv5_6/x2/bn_w_0"], (ndims(weights["conv5_6/x2/bn_w_0"]) + 1) - 1)
edge_200 = (Flux.unsqueeze)(weights["conv5_6/x2/bn_b_0"], (ndims(weights["conv5_6/x2/bn_b_0"]) + 1) - 1)
edge_201 = (Flux.unsqueeze)(weights["conv5_7/x1/bn_w_0"], (ndims(weights["conv5_7/x1/bn_w_0"]) + 1) - 1)
edge_202 = (Flux.unsqueeze)(weights["conv5_7/x1/bn_b_0"], (ndims(weights["conv5_7/x1/bn_b_0"]) + 1) - 1)
edge_203 = (Flux.unsqueeze)(weights["conv5_7/x2/bn_w_0"], (ndims(weights["conv5_7/x2/bn_w_0"]) + 1) - 1)
edge_204 = (Flux.unsqueeze)(weights["conv5_7/x2/bn_b_0"], (ndims(weights["conv5_7/x2/bn_b_0"]) + 1) - 1)
edge_205 = (Flux.unsqueeze)(weights["conv5_8/x1/bn_w_0"], (ndims(weights["conv5_8/x1/bn_w_0"]) + 1) - 1)
edge_206 = (Flux.unsqueeze)(weights["conv5_8/x1/bn_b_0"], (ndims(weights["conv5_8/x1/bn_b_0"]) + 1) - 1)
edge_207 = (Flux.unsqueeze)(weights["conv5_8/x2/bn_w_0"], (ndims(weights["conv5_8/x2/bn_w_0"]) + 1) - 1)
edge_208 = (Flux.unsqueeze)(weights["conv5_8/x2/bn_b_0"], (ndims(weights["conv5_8/x2/bn_b_0"]) + 1) - 1)
edge_209 = (Flux.unsqueeze)(weights["conv5_9/x1/bn_w_0"], (ndims(weights["conv5_9/x1/bn_w_0"]) + 1) - 1)
edge_210 = (Flux.unsqueeze)(weights["conv5_9/x1/bn_b_0"], (ndims(weights["conv5_9/x1/bn_b_0"]) + 1) - 1)
edge_211 = (Flux.unsqueeze)(weights["conv5_9/x2/bn_w_0"], (ndims(weights["conv5_9/x2/bn_w_0"]) + 1) - 1)
edge_212 = (Flux.unsqueeze)(weights["conv5_9/x2/bn_b_0"], (ndims(weights["conv5_9/x2/bn_b_0"]) + 1) - 1)
edge_213 = (Flux.unsqueeze)(weights["conv5_10/x1/bn_w_0"], (ndims(weights["conv5_10/x1/bn_w_0"]) + 1) - 1)
edge_214 = (Flux.unsqueeze)(weights["conv5_10/x1/bn_b_0"], (ndims(weights["conv5_10/x1/bn_b_0"]) + 1) - 1)
edge_215 = (Flux.unsqueeze)(weights["conv5_10/x2/bn_w_0"], (ndims(weights["conv5_10/x2/bn_w_0"]) + 1) - 1)
edge_216 = (Flux.unsqueeze)(weights["conv5_10/x2/bn_b_0"], (ndims(weights["conv5_10/x2/bn_b_0"]) + 1) - 1)
edge_217 = (Flux.unsqueeze)(weights["conv5_11/x1/bn_w_0"], (ndims(weights["conv5_11/x1/bn_w_0"]) + 1) - 1)
edge_218 = (Flux.unsqueeze)(weights["conv5_11/x1/bn_b_0"], (ndims(weights["conv5_11/x1/bn_b_0"]) + 1) - 1)
edge_219 = (Flux.unsqueeze)(weights["conv5_11/x2/bn_w_0"], (ndims(weights["conv5_11/x2/bn_w_0"]) + 1) - 1)
edge_220 = (Flux.unsqueeze)(weights["conv5_11/x2/bn_b_0"], (ndims(weights["conv5_11/x2/bn_b_0"]) + 1) - 1)
edge_221 = (Flux.unsqueeze)(weights["conv5_12/x1/bn_w_0"], (ndims(weights["conv5_12/x1/bn_w_0"]) + 1) - 1)
edge_222 = (Flux.unsqueeze)(weights["conv5_12/x1/bn_b_0"], (ndims(weights["conv5_12/x1/bn_b_0"]) + 1) - 1)
edge_223 = (Flux.unsqueeze)(weights["conv5_12/x2/bn_w_0"], (ndims(weights["conv5_12/x2/bn_w_0"]) + 1) - 1)
edge_224 = (Flux.unsqueeze)(weights["conv5_12/x2/bn_b_0"], (ndims(weights["conv5_12/x2/bn_b_0"]) + 1) - 1)
edge_225 = (Flux.unsqueeze)(weights["conv5_13/x1/bn_w_0"], (ndims(weights["conv5_13/x1/bn_w_0"]) + 1) - 1)
edge_226 = (Flux.unsqueeze)(weights["conv5_13/x1/bn_b_0"], (ndims(weights["conv5_13/x1/bn_b_0"]) + 1) - 1)
edge_227 = (Flux.unsqueeze)(weights["conv5_13/x2/bn_w_0"], (ndims(weights["conv5_13/x2/bn_w_0"]) + 1) - 1)
edge_228 = (Flux.unsqueeze)(weights["conv5_13/x2/bn_b_0"], (ndims(weights["conv5_13/x2/bn_b_0"]) + 1) - 1)
edge_229 = (Flux.unsqueeze)(weights["conv5_14/x1/bn_w_0"], (ndims(weights["conv5_14/x1/bn_w_0"]) + 1) - 1)
edge_230 = (Flux.unsqueeze)(weights["conv5_14/x1/bn_b_0"], (ndims(weights["conv5_14/x1/bn_b_0"]) + 1) - 1)
edge_231 = (Flux.unsqueeze)(weights["conv5_14/x2/bn_w_0"], (ndims(weights["conv5_14/x2/bn_w_0"]) + 1) - 1)
edge_232 = (Flux.unsqueeze)(weights["conv5_14/x2/bn_b_0"], (ndims(weights["conv5_14/x2/bn_b_0"]) + 1) - 1)
edge_233 = (Flux.unsqueeze)(weights["conv5_15/x1/bn_w_0"], (ndims(weights["conv5_15/x1/bn_w_0"]) + 1) - 1)
edge_234 = (Flux.unsqueeze)(weights["conv5_15/x1/bn_b_0"], (ndims(weights["conv5_15/x1/bn_b_0"]) + 1) - 1)
edge_235 = (Flux.unsqueeze)(weights["conv5_15/x2/bn_w_0"], (ndims(weights["conv5_15/x2/bn_w_0"]) + 1) - 1)
edge_236 = (Flux.unsqueeze)(weights["conv5_15/x2/bn_b_0"], (ndims(weights["conv5_15/x2/bn_b_0"]) + 1) - 1)
edge_237 = (Flux.unsqueeze)(weights["conv5_16/x1/bn_w_0"], (ndims(weights["conv5_16/x1/bn_w_0"]) + 1) - 1)
edge_238 = (Flux.unsqueeze)(weights["conv5_16/x1/bn_b_0"], (ndims(weights["conv5_16/x1/bn_b_0"]) + 1) - 1)
edge_239 = (Flux.unsqueeze)(weights["conv5_16/x2/bn_w_0"], (ndims(weights["conv5_16/x2/bn_w_0"]) + 1) - 1)
edge_240 = (Flux.unsqueeze)(weights["conv5_16/x2/bn_b_0"], (ndims(weights["conv5_16/x2/bn_b_0"]) + 1) - 1)
edge_241 = (Flux.unsqueeze)(weights["conv5_blk/bn_w_0"], (ndims(weights["conv5_blk/bn_w_0"]) + 1) - 1)
edge_242 = (Flux.unsqueeze)(weights["conv5_blk/bn_b_0"], (ndims(weights["conv5_blk/bn_b_0"]) + 1) - 1)
c_243 = Conv(flipkernel(weights["fc6_w_0"]), weights["fc6_b_0"], stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_244 = BatchNorm(identity, weights["conv5_blk/bn_bias_0"], weights["conv5_blk/bn_scale_0"], weights["conv5_blk/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_blk/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_245 = Conv(flipkernel(weights["conv4_blk_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_246 = BatchNorm(identity, weights["conv4_blk/bn_bias_0"], weights["conv4_blk/bn_scale_0"], weights["conv4_blk/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_blk/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_247 = Conv(flipkernel(weights["conv3_blk_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_248 = BatchNorm(identity, weights["conv3_blk/bn_bias_0"], weights["conv3_blk/bn_scale_0"], weights["conv3_blk/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_blk/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_249 = Conv(flipkernel(weights["conv2_blk_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_250 = BatchNorm(identity, weights["conv2_blk/bn_bias_0"], weights["conv2_blk/bn_scale_0"], weights["conv2_blk/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_blk/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_251 = BatchNorm(identity, weights["conv1/bn_bias_0"], weights["conv1/bn_scale_0"], weights["conv1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_252 = Conv(flipkernel(weights["conv1_w_0"]), Float32[0.0], relu, stride=(2, 2), pad=(3, 3), dilation=(1, 1))
c_253 = (Flux.unsqueeze)(edge_1, (ndims(edge_1) + 1) - 2)
c_254 = (Flux.unsqueeze)(edge_2, (ndims(edge_2) + 1) - 2)
c_255 = Conv(flipkernel(weights["conv2_1/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_256 = BatchNorm(identity, weights["conv2_1/x2/bn_bias_0"], weights["conv2_1/x2/bn_scale_0"], weights["conv2_1/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_1/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_257 = Conv(flipkernel(weights["conv2_1/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_258 = BatchNorm(identity, weights["conv2_1/x1/bn_bias_0"], weights["conv2_1/x1/bn_scale_0"], weights["conv2_1/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_1/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_259 = (Flux.unsqueeze)(edge_3, (ndims(edge_3) + 1) - 2)
c_260 = (Flux.unsqueeze)(edge_4, (ndims(edge_4) + 1) - 2)
c_261 = (Flux.unsqueeze)(edge_5, (ndims(edge_5) + 1) - 2)
c_262 = (Flux.unsqueeze)(edge_6, (ndims(edge_6) + 1) - 2)
c_263 = Conv(flipkernel(weights["conv2_2/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_264 = BatchNorm(identity, weights["conv2_2/x2/bn_bias_0"], weights["conv2_2/x2/bn_scale_0"], weights["conv2_2/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_2/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_265 = Conv(flipkernel(weights["conv2_2/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_266 = BatchNorm(identity, weights["conv2_2/x1/bn_bias_0"], weights["conv2_2/x1/bn_scale_0"], weights["conv2_2/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_2/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_267 = (Flux.unsqueeze)(edge_7, (ndims(edge_7) + 1) - 2)
c_268 = (Flux.unsqueeze)(edge_8, (ndims(edge_8) + 1) - 2)
c_269 = (Flux.unsqueeze)(edge_9, (ndims(edge_9) + 1) - 2)
c_270 = (Flux.unsqueeze)(edge_10, (ndims(edge_10) + 1) - 2)
c_271 = Conv(flipkernel(weights["conv2_3/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_272 = BatchNorm(identity, weights["conv2_3/x2/bn_bias_0"], weights["conv2_3/x2/bn_scale_0"], weights["conv2_3/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_3/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_273 = Conv(flipkernel(weights["conv2_3/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_274 = BatchNorm(identity, weights["conv2_3/x1/bn_bias_0"], weights["conv2_3/x1/bn_scale_0"], weights["conv2_3/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_3/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_275 = (Flux.unsqueeze)(edge_11, (ndims(edge_11) + 1) - 2)
c_276 = (Flux.unsqueeze)(edge_12, (ndims(edge_12) + 1) - 2)
c_277 = (Flux.unsqueeze)(edge_13, (ndims(edge_13) + 1) - 2)
c_278 = (Flux.unsqueeze)(edge_14, (ndims(edge_14) + 1) - 2)
c_279 = Conv(flipkernel(weights["conv2_4/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_280 = BatchNorm(identity, weights["conv2_4/x2/bn_bias_0"], weights["conv2_4/x2/bn_scale_0"], weights["conv2_4/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_4/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_281 = Conv(flipkernel(weights["conv2_4/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_282 = BatchNorm(identity, weights["conv2_4/x1/bn_bias_0"], weights["conv2_4/x1/bn_scale_0"], weights["conv2_4/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_4/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_283 = (Flux.unsqueeze)(edge_15, (ndims(edge_15) + 1) - 2)
c_284 = (Flux.unsqueeze)(edge_16, (ndims(edge_16) + 1) - 2)
c_285 = (Flux.unsqueeze)(edge_17, (ndims(edge_17) + 1) - 2)
c_286 = (Flux.unsqueeze)(edge_18, (ndims(edge_18) + 1) - 2)
c_287 = Conv(flipkernel(weights["conv2_5/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_288 = BatchNorm(identity, weights["conv2_5/x2/bn_bias_0"], weights["conv2_5/x2/bn_scale_0"], weights["conv2_5/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_5/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_289 = Conv(flipkernel(weights["conv2_5/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_290 = BatchNorm(identity, weights["conv2_5/x1/bn_bias_0"], weights["conv2_5/x1/bn_scale_0"], weights["conv2_5/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_5/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_291 = (Flux.unsqueeze)(edge_19, (ndims(edge_19) + 1) - 2)
c_292 = (Flux.unsqueeze)(edge_20, (ndims(edge_20) + 1) - 2)
c_293 = (Flux.unsqueeze)(edge_21, (ndims(edge_21) + 1) - 2)
c_294 = (Flux.unsqueeze)(edge_22, (ndims(edge_22) + 1) - 2)
c_295 = Conv(flipkernel(weights["conv2_6/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_296 = BatchNorm(identity, weights["conv2_6/x2/bn_bias_0"], weights["conv2_6/x2/bn_scale_0"], weights["conv2_6/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_6/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_297 = Conv(flipkernel(weights["conv2_6/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_298 = BatchNorm(identity, weights["conv2_6/x1/bn_bias_0"], weights["conv2_6/x1/bn_scale_0"], weights["conv2_6/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv2_6/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_299 = (Flux.unsqueeze)(edge_23, (ndims(edge_23) + 1) - 2)
c_300 = (Flux.unsqueeze)(edge_24, (ndims(edge_24) + 1) - 2)
c_301 = (Flux.unsqueeze)(edge_25, (ndims(edge_25) + 1) - 2)
c_302 = (Flux.unsqueeze)(edge_26, (ndims(edge_26) + 1) - 2)
c_303 = (Flux.unsqueeze)(edge_27, (ndims(edge_27) + 1) - 2)
c_304 = (Flux.unsqueeze)(edge_28, (ndims(edge_28) + 1) - 2)
c_305 = Conv(flipkernel(weights["conv3_1/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_306 = BatchNorm(identity, weights["conv3_1/x2/bn_bias_0"], weights["conv3_1/x2/bn_scale_0"], weights["conv3_1/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_1/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_307 = Conv(flipkernel(weights["conv3_1/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_308 = BatchNorm(identity, weights["conv3_1/x1/bn_bias_0"], weights["conv3_1/x1/bn_scale_0"], weights["conv3_1/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_1/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_309 = (Flux.unsqueeze)(edge_29, (ndims(edge_29) + 1) - 2)
c_310 = (Flux.unsqueeze)(edge_30, (ndims(edge_30) + 1) - 2)
c_311 = (Flux.unsqueeze)(edge_31, (ndims(edge_31) + 1) - 2)
c_312 = (Flux.unsqueeze)(edge_32, (ndims(edge_32) + 1) - 2)
c_313 = Conv(flipkernel(weights["conv3_2/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_314 = BatchNorm(identity, weights["conv3_2/x2/bn_bias_0"], weights["conv3_2/x2/bn_scale_0"], weights["conv3_2/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_2/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_315 = Conv(flipkernel(weights["conv3_2/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_316 = BatchNorm(identity, weights["conv3_2/x1/bn_bias_0"], weights["conv3_2/x1/bn_scale_0"], weights["conv3_2/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_2/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_317 = (Flux.unsqueeze)(edge_33, (ndims(edge_33) + 1) - 2)
c_318 = (Flux.unsqueeze)(edge_34, (ndims(edge_34) + 1) - 2)
c_319 = (Flux.unsqueeze)(edge_35, (ndims(edge_35) + 1) - 2)
c_320 = (Flux.unsqueeze)(edge_36, (ndims(edge_36) + 1) - 2)
c_321 = Conv(flipkernel(weights["conv3_3/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_322 = BatchNorm(identity, weights["conv3_3/x2/bn_bias_0"], weights["conv3_3/x2/bn_scale_0"], weights["conv3_3/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_3/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_323 = Conv(flipkernel(weights["conv3_3/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_324 = BatchNorm(identity, weights["conv3_3/x1/bn_bias_0"], weights["conv3_3/x1/bn_scale_0"], weights["conv3_3/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_3/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_325 = (Flux.unsqueeze)(edge_37, (ndims(edge_37) + 1) - 2)
c_326 = (Flux.unsqueeze)(edge_38, (ndims(edge_38) + 1) - 2)
c_327 = (Flux.unsqueeze)(edge_39, (ndims(edge_39) + 1) - 2)
c_328 = (Flux.unsqueeze)(edge_40, (ndims(edge_40) + 1) - 2)
c_329 = Conv(flipkernel(weights["conv3_4/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_330 = BatchNorm(identity, weights["conv3_4/x2/bn_bias_0"], weights["conv3_4/x2/bn_scale_0"], weights["conv3_4/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_4/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_331 = Conv(flipkernel(weights["conv3_4/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_332 = BatchNorm(identity, weights["conv3_4/x1/bn_bias_0"], weights["conv3_4/x1/bn_scale_0"], weights["conv3_4/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_4/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_333 = (Flux.unsqueeze)(edge_41, (ndims(edge_41) + 1) - 2)
c_334 = (Flux.unsqueeze)(edge_42, (ndims(edge_42) + 1) - 2)
c_335 = (Flux.unsqueeze)(edge_43, (ndims(edge_43) + 1) - 2)
c_336 = (Flux.unsqueeze)(edge_44, (ndims(edge_44) + 1) - 2)
c_337 = Conv(flipkernel(weights["conv3_5/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_338 = BatchNorm(identity, weights["conv3_5/x2/bn_bias_0"], weights["conv3_5/x2/bn_scale_0"], weights["conv3_5/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_5/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_339 = Conv(flipkernel(weights["conv3_5/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_340 = BatchNorm(identity, weights["conv3_5/x1/bn_bias_0"], weights["conv3_5/x1/bn_scale_0"], weights["conv3_5/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_5/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_341 = (Flux.unsqueeze)(edge_45, (ndims(edge_45) + 1) - 2)
c_342 = (Flux.unsqueeze)(edge_46, (ndims(edge_46) + 1) - 2)
c_343 = (Flux.unsqueeze)(edge_47, (ndims(edge_47) + 1) - 2)
c_344 = (Flux.unsqueeze)(edge_48, (ndims(edge_48) + 1) - 2)
c_345 = Conv(flipkernel(weights["conv3_6/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_346 = BatchNorm(identity, weights["conv3_6/x2/bn_bias_0"], weights["conv3_6/x2/bn_scale_0"], weights["conv3_6/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_6/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_347 = Conv(flipkernel(weights["conv3_6/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_348 = BatchNorm(identity, weights["conv3_6/x1/bn_bias_0"], weights["conv3_6/x1/bn_scale_0"], weights["conv3_6/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_6/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_349 = (Flux.unsqueeze)(edge_49, (ndims(edge_49) + 1) - 2)
c_350 = (Flux.unsqueeze)(edge_50, (ndims(edge_50) + 1) - 2)
c_351 = (Flux.unsqueeze)(edge_51, (ndims(edge_51) + 1) - 2)
c_352 = (Flux.unsqueeze)(edge_52, (ndims(edge_52) + 1) - 2)
c_353 = Conv(flipkernel(weights["conv3_7/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_354 = BatchNorm(identity, weights["conv3_7/x2/bn_bias_0"], weights["conv3_7/x2/bn_scale_0"], weights["conv3_7/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_7/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_355 = Conv(flipkernel(weights["conv3_7/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_356 = BatchNorm(identity, weights["conv3_7/x1/bn_bias_0"], weights["conv3_7/x1/bn_scale_0"], weights["conv3_7/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_7/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_357 = (Flux.unsqueeze)(edge_53, (ndims(edge_53) + 1) - 2)
c_358 = (Flux.unsqueeze)(edge_54, (ndims(edge_54) + 1) - 2)
c_359 = (Flux.unsqueeze)(edge_55, (ndims(edge_55) + 1) - 2)
c_360 = (Flux.unsqueeze)(edge_56, (ndims(edge_56) + 1) - 2)
c_361 = Conv(flipkernel(weights["conv3_8/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_362 = BatchNorm(identity, weights["conv3_8/x2/bn_bias_0"], weights["conv3_8/x2/bn_scale_0"], weights["conv3_8/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_8/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_363 = Conv(flipkernel(weights["conv3_8/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_364 = BatchNorm(identity, weights["conv3_8/x1/bn_bias_0"], weights["conv3_8/x1/bn_scale_0"], weights["conv3_8/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_8/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_365 = (Flux.unsqueeze)(edge_57, (ndims(edge_57) + 1) - 2)
c_366 = (Flux.unsqueeze)(edge_58, (ndims(edge_58) + 1) - 2)
c_367 = (Flux.unsqueeze)(edge_59, (ndims(edge_59) + 1) - 2)
c_368 = (Flux.unsqueeze)(edge_60, (ndims(edge_60) + 1) - 2)
c_369 = Conv(flipkernel(weights["conv3_9/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_370 = BatchNorm(identity, weights["conv3_9/x2/bn_bias_0"], weights["conv3_9/x2/bn_scale_0"], weights["conv3_9/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_9/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_371 = Conv(flipkernel(weights["conv3_9/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_372 = BatchNorm(identity, weights["conv3_9/x1/bn_bias_0"], weights["conv3_9/x1/bn_scale_0"], weights["conv3_9/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_9/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_373 = (Flux.unsqueeze)(edge_61, (ndims(edge_61) + 1) - 2)
c_374 = (Flux.unsqueeze)(edge_62, (ndims(edge_62) + 1) - 2)
c_375 = (Flux.unsqueeze)(edge_63, (ndims(edge_63) + 1) - 2)
c_376 = (Flux.unsqueeze)(edge_64, (ndims(edge_64) + 1) - 2)
c_377 = Conv(flipkernel(weights["conv3_10/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_378 = BatchNorm(identity, weights["conv3_10/x2/bn_bias_0"], weights["conv3_10/x2/bn_scale_0"], weights["conv3_10/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_10/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_379 = Conv(flipkernel(weights["conv3_10/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_380 = BatchNorm(identity, weights["conv3_10/x1/bn_bias_0"], weights["conv3_10/x1/bn_scale_0"], weights["conv3_10/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_10/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_381 = (Flux.unsqueeze)(edge_65, (ndims(edge_65) + 1) - 2)
c_382 = (Flux.unsqueeze)(edge_66, (ndims(edge_66) + 1) - 2)
c_383 = (Flux.unsqueeze)(edge_67, (ndims(edge_67) + 1) - 2)
c_384 = (Flux.unsqueeze)(edge_68, (ndims(edge_68) + 1) - 2)
c_385 = Conv(flipkernel(weights["conv3_11/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_386 = BatchNorm(identity, weights["conv3_11/x2/bn_bias_0"], weights["conv3_11/x2/bn_scale_0"], weights["conv3_11/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_11/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_387 = Conv(flipkernel(weights["conv3_11/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_388 = BatchNorm(identity, weights["conv3_11/x1/bn_bias_0"], weights["conv3_11/x1/bn_scale_0"], weights["conv3_11/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_11/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_389 = (Flux.unsqueeze)(edge_69, (ndims(edge_69) + 1) - 2)
c_390 = (Flux.unsqueeze)(edge_70, (ndims(edge_70) + 1) - 2)
c_391 = (Flux.unsqueeze)(edge_71, (ndims(edge_71) + 1) - 2)
c_392 = (Flux.unsqueeze)(edge_72, (ndims(edge_72) + 1) - 2)
c_393 = Conv(flipkernel(weights["conv3_12/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_394 = BatchNorm(identity, weights["conv3_12/x2/bn_bias_0"], weights["conv3_12/x2/bn_scale_0"], weights["conv3_12/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_12/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_395 = Conv(flipkernel(weights["conv3_12/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_396 = BatchNorm(identity, weights["conv3_12/x1/bn_bias_0"], weights["conv3_12/x1/bn_scale_0"], weights["conv3_12/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv3_12/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_397 = (Flux.unsqueeze)(edge_73, (ndims(edge_73) + 1) - 2)
c_398 = (Flux.unsqueeze)(edge_74, (ndims(edge_74) + 1) - 2)
c_399 = (Flux.unsqueeze)(edge_75, (ndims(edge_75) + 1) - 2)
c_400 = (Flux.unsqueeze)(edge_76, (ndims(edge_76) + 1) - 2)
c_401 = (Flux.unsqueeze)(edge_77, (ndims(edge_77) + 1) - 2)
c_402 = (Flux.unsqueeze)(edge_78, (ndims(edge_78) + 1) - 2)
c_403 = Conv(flipkernel(weights["conv4_1/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_404 = BatchNorm(identity, weights["conv4_1/x2/bn_bias_0"], weights["conv4_1/x2/bn_scale_0"], weights["conv4_1/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_1/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_405 = Conv(flipkernel(weights["conv4_1/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_406 = BatchNorm(identity, weights["conv4_1/x1/bn_bias_0"], weights["conv4_1/x1/bn_scale_0"], weights["conv4_1/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_1/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_407 = (Flux.unsqueeze)(edge_79, (ndims(edge_79) + 1) - 2)
c_408 = (Flux.unsqueeze)(edge_80, (ndims(edge_80) + 1) - 2)
c_409 = (Flux.unsqueeze)(edge_81, (ndims(edge_81) + 1) - 2)
c_410 = (Flux.unsqueeze)(edge_82, (ndims(edge_82) + 1) - 2)
c_411 = Conv(flipkernel(weights["conv4_2/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_412 = BatchNorm(identity, weights["conv4_2/x2/bn_bias_0"], weights["conv4_2/x2/bn_scale_0"], weights["conv4_2/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_2/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_413 = Conv(flipkernel(weights["conv4_2/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_414 = BatchNorm(identity, weights["conv4_2/x1/bn_bias_0"], weights["conv4_2/x1/bn_scale_0"], weights["conv4_2/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_2/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_415 = (Flux.unsqueeze)(edge_83, (ndims(edge_83) + 1) - 2)
c_416 = (Flux.unsqueeze)(edge_84, (ndims(edge_84) + 1) - 2)
c_417 = (Flux.unsqueeze)(edge_85, (ndims(edge_85) + 1) - 2)
c_418 = (Flux.unsqueeze)(edge_86, (ndims(edge_86) + 1) - 2)
c_419 = Conv(flipkernel(weights["conv4_3/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_420 = BatchNorm(identity, weights["conv4_3/x2/bn_bias_0"], weights["conv4_3/x2/bn_scale_0"], weights["conv4_3/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_3/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_421 = Conv(flipkernel(weights["conv4_3/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_422 = BatchNorm(identity, weights["conv4_3/x1/bn_bias_0"], weights["conv4_3/x1/bn_scale_0"], weights["conv4_3/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_3/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_423 = (Flux.unsqueeze)(edge_87, (ndims(edge_87) + 1) - 2)
c_424 = (Flux.unsqueeze)(edge_88, (ndims(edge_88) + 1) - 2)
c_425 = (Flux.unsqueeze)(edge_89, (ndims(edge_89) + 1) - 2)
c_426 = (Flux.unsqueeze)(edge_90, (ndims(edge_90) + 1) - 2)
c_427 = Conv(flipkernel(weights["conv4_4/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_428 = BatchNorm(identity, weights["conv4_4/x2/bn_bias_0"], weights["conv4_4/x2/bn_scale_0"], weights["conv4_4/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_4/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_429 = Conv(flipkernel(weights["conv4_4/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_430 = BatchNorm(identity, weights["conv4_4/x1/bn_bias_0"], weights["conv4_4/x1/bn_scale_0"], weights["conv4_4/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_4/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_431 = (Flux.unsqueeze)(edge_91, (ndims(edge_91) + 1) - 2)
c_432 = (Flux.unsqueeze)(edge_92, (ndims(edge_92) + 1) - 2)
c_433 = (Flux.unsqueeze)(edge_93, (ndims(edge_93) + 1) - 2)
c_434 = (Flux.unsqueeze)(edge_94, (ndims(edge_94) + 1) - 2)
c_435 = Conv(flipkernel(weights["conv4_5/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_436 = BatchNorm(identity, weights["conv4_5/x2/bn_bias_0"], weights["conv4_5/x2/bn_scale_0"], weights["conv4_5/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_5/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_437 = Conv(flipkernel(weights["conv4_5/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_438 = BatchNorm(identity, weights["conv4_5/x1/bn_bias_0"], weights["conv4_5/x1/bn_scale_0"], weights["conv4_5/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_5/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_439 = (Flux.unsqueeze)(edge_95, (ndims(edge_95) + 1) - 2)
c_440 = (Flux.unsqueeze)(edge_96, (ndims(edge_96) + 1) - 2)
c_441 = (Flux.unsqueeze)(edge_97, (ndims(edge_97) + 1) - 2)
c_442 = (Flux.unsqueeze)(edge_98, (ndims(edge_98) + 1) - 2)
c_443 = Conv(flipkernel(weights["conv4_6/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_444 = BatchNorm(identity, weights["conv4_6/x2/bn_bias_0"], weights["conv4_6/x2/bn_scale_0"], weights["conv4_6/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_6/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_445 = Conv(flipkernel(weights["conv4_6/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_446 = BatchNorm(identity, weights["conv4_6/x1/bn_bias_0"], weights["conv4_6/x1/bn_scale_0"], weights["conv4_6/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_6/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_447 = (Flux.unsqueeze)(edge_99, (ndims(edge_99) + 1) - 2)
c_448 = (Flux.unsqueeze)(edge_100, (ndims(edge_100) + 1) - 2)
c_449 = (Flux.unsqueeze)(edge_101, (ndims(edge_101) + 1) - 2)
c_450 = (Flux.unsqueeze)(edge_102, (ndims(edge_102) + 1) - 2)
c_451 = Conv(flipkernel(weights["conv4_7/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_452 = BatchNorm(identity, weights["conv4_7/x2/bn_bias_0"], weights["conv4_7/x2/bn_scale_0"], weights["conv4_7/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_7/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_453 = Conv(flipkernel(weights["conv4_7/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_454 = BatchNorm(identity, weights["conv4_7/x1/bn_bias_0"], weights["conv4_7/x1/bn_scale_0"], weights["conv4_7/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_7/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_455 = (Flux.unsqueeze)(edge_103, (ndims(edge_103) + 1) - 2)
c_456 = (Flux.unsqueeze)(edge_104, (ndims(edge_104) + 1) - 2)
c_457 = (Flux.unsqueeze)(edge_105, (ndims(edge_105) + 1) - 2)
c_458 = (Flux.unsqueeze)(edge_106, (ndims(edge_106) + 1) - 2)
c_459 = Conv(flipkernel(weights["conv4_8/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_460 = BatchNorm(identity, weights["conv4_8/x2/bn_bias_0"], weights["conv4_8/x2/bn_scale_0"], weights["conv4_8/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_8/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_461 = Conv(flipkernel(weights["conv4_8/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_462 = BatchNorm(identity, weights["conv4_8/x1/bn_bias_0"], weights["conv4_8/x1/bn_scale_0"], weights["conv4_8/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_8/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_463 = (Flux.unsqueeze)(edge_107, (ndims(edge_107) + 1) - 2)
c_464 = (Flux.unsqueeze)(edge_108, (ndims(edge_108) + 1) - 2)
c_465 = (Flux.unsqueeze)(edge_109, (ndims(edge_109) + 1) - 2)
c_466 = (Flux.unsqueeze)(edge_110, (ndims(edge_110) + 1) - 2)
c_467 = Conv(flipkernel(weights["conv4_9/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_468 = BatchNorm(identity, weights["conv4_9/x2/bn_bias_0"], weights["conv4_9/x2/bn_scale_0"], weights["conv4_9/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_9/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_469 = Conv(flipkernel(weights["conv4_9/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_470 = BatchNorm(identity, weights["conv4_9/x1/bn_bias_0"], weights["conv4_9/x1/bn_scale_0"], weights["conv4_9/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_9/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_471 = (Flux.unsqueeze)(edge_111, (ndims(edge_111) + 1) - 2)
c_472 = (Flux.unsqueeze)(edge_112, (ndims(edge_112) + 1) - 2)
c_473 = (Flux.unsqueeze)(edge_113, (ndims(edge_113) + 1) - 2)
c_474 = (Flux.unsqueeze)(edge_114, (ndims(edge_114) + 1) - 2)
c_475 = Conv(flipkernel(weights["conv4_10/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_476 = BatchNorm(identity, weights["conv4_10/x2/bn_bias_0"], weights["conv4_10/x2/bn_scale_0"], weights["conv4_10/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_10/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_477 = Conv(flipkernel(weights["conv4_10/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_478 = BatchNorm(identity, weights["conv4_10/x1/bn_bias_0"], weights["conv4_10/x1/bn_scale_0"], weights["conv4_10/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_10/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_479 = (Flux.unsqueeze)(edge_115, (ndims(edge_115) + 1) - 2)
c_480 = (Flux.unsqueeze)(edge_116, (ndims(edge_116) + 1) - 2)
c_481 = (Flux.unsqueeze)(edge_117, (ndims(edge_117) + 1) - 2)
c_482 = (Flux.unsqueeze)(edge_118, (ndims(edge_118) + 1) - 2)
c_483 = Conv(flipkernel(weights["conv4_11/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_484 = BatchNorm(identity, weights["conv4_11/x2/bn_bias_0"], weights["conv4_11/x2/bn_scale_0"], weights["conv4_11/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_11/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_485 = Conv(flipkernel(weights["conv4_11/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_486 = BatchNorm(identity, weights["conv4_11/x1/bn_bias_0"], weights["conv4_11/x1/bn_scale_0"], weights["conv4_11/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_11/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_487 = (Flux.unsqueeze)(edge_119, (ndims(edge_119) + 1) - 2)
c_488 = (Flux.unsqueeze)(edge_120, (ndims(edge_120) + 1) - 2)
c_489 = (Flux.unsqueeze)(edge_121, (ndims(edge_121) + 1) - 2)
c_490 = (Flux.unsqueeze)(edge_122, (ndims(edge_122) + 1) - 2)
c_491 = Conv(flipkernel(weights["conv4_12/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_492 = BatchNorm(identity, weights["conv4_12/x2/bn_bias_0"], weights["conv4_12/x2/bn_scale_0"], weights["conv4_12/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_12/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_493 = Conv(flipkernel(weights["conv4_12/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_494 = BatchNorm(identity, weights["conv4_12/x1/bn_bias_0"], weights["conv4_12/x1/bn_scale_0"], weights["conv4_12/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_12/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_495 = (Flux.unsqueeze)(edge_123, (ndims(edge_123) + 1) - 2)
c_496 = (Flux.unsqueeze)(edge_124, (ndims(edge_124) + 1) - 2)
c_497 = (Flux.unsqueeze)(edge_125, (ndims(edge_125) + 1) - 2)
c_498 = (Flux.unsqueeze)(edge_126, (ndims(edge_126) + 1) - 2)
c_499 = Conv(flipkernel(weights["conv4_13/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_500 = BatchNorm(identity, weights["conv4_13/x2/bn_bias_0"], weights["conv4_13/x2/bn_scale_0"], weights["conv4_13/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_13/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_501 = Conv(flipkernel(weights["conv4_13/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_502 = BatchNorm(identity, weights["conv4_13/x1/bn_bias_0"], weights["conv4_13/x1/bn_scale_0"], weights["conv4_13/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_13/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_503 = (Flux.unsqueeze)(edge_127, (ndims(edge_127) + 1) - 2)
c_504 = (Flux.unsqueeze)(edge_128, (ndims(edge_128) + 1) - 2)
c_505 = (Flux.unsqueeze)(edge_129, (ndims(edge_129) + 1) - 2)
c_506 = (Flux.unsqueeze)(edge_130, (ndims(edge_130) + 1) - 2)
c_507 = Conv(flipkernel(weights["conv4_14/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_508 = BatchNorm(identity, weights["conv4_14/x2/bn_bias_0"], weights["conv4_14/x2/bn_scale_0"], weights["conv4_14/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_14/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_509 = Conv(flipkernel(weights["conv4_14/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_510 = BatchNorm(identity, weights["conv4_14/x1/bn_bias_0"], weights["conv4_14/x1/bn_scale_0"], weights["conv4_14/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_14/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_511 = (Flux.unsqueeze)(edge_131, (ndims(edge_131) + 1) - 2)
c_512 = (Flux.unsqueeze)(edge_132, (ndims(edge_132) + 1) - 2)
c_513 = (Flux.unsqueeze)(edge_133, (ndims(edge_133) + 1) - 2)
c_514 = (Flux.unsqueeze)(edge_134, (ndims(edge_134) + 1) - 2)
c_515 = Conv(flipkernel(weights["conv4_15/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_516 = BatchNorm(identity, weights["conv4_15/x2/bn_bias_0"], weights["conv4_15/x2/bn_scale_0"], weights["conv4_15/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_15/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_517 = Conv(flipkernel(weights["conv4_15/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_518 = BatchNorm(identity, weights["conv4_15/x1/bn_bias_0"], weights["conv4_15/x1/bn_scale_0"], weights["conv4_15/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_15/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_519 = (Flux.unsqueeze)(edge_135, (ndims(edge_135) + 1) - 2)
c_520 = (Flux.unsqueeze)(edge_136, (ndims(edge_136) + 1) - 2)
c_521 = (Flux.unsqueeze)(edge_137, (ndims(edge_137) + 1) - 2)
c_522 = (Flux.unsqueeze)(edge_138, (ndims(edge_138) + 1) - 2)
c_523 = Conv(flipkernel(weights["conv4_16/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_524 = BatchNorm(identity, weights["conv4_16/x2/bn_bias_0"], weights["conv4_16/x2/bn_scale_0"], weights["conv4_16/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_16/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_525 = Conv(flipkernel(weights["conv4_16/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_526 = BatchNorm(identity, weights["conv4_16/x1/bn_bias_0"], weights["conv4_16/x1/bn_scale_0"], weights["conv4_16/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_16/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_527 = (Flux.unsqueeze)(edge_139, (ndims(edge_139) + 1) - 2)
c_528 = (Flux.unsqueeze)(edge_140, (ndims(edge_140) + 1) - 2)
c_529 = (Flux.unsqueeze)(edge_141, (ndims(edge_141) + 1) - 2)
c_530 = (Flux.unsqueeze)(edge_142, (ndims(edge_142) + 1) - 2)
c_531 = Conv(flipkernel(weights["conv4_17/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_532 = BatchNorm(identity, weights["conv4_17/x2/bn_bias_0"], weights["conv4_17/x2/bn_scale_0"], weights["conv4_17/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_17/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_533 = Conv(flipkernel(weights["conv4_17/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_534 = BatchNorm(identity, weights["conv4_17/x1/bn_bias_0"], weights["conv4_17/x1/bn_scale_0"], weights["conv4_17/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_17/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_535 = (Flux.unsqueeze)(edge_143, (ndims(edge_143) + 1) - 2)
c_536 = (Flux.unsqueeze)(edge_144, (ndims(edge_144) + 1) - 2)
c_537 = (Flux.unsqueeze)(edge_145, (ndims(edge_145) + 1) - 2)
c_538 = (Flux.unsqueeze)(edge_146, (ndims(edge_146) + 1) - 2)
c_539 = Conv(flipkernel(weights["conv4_18/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_540 = BatchNorm(identity, weights["conv4_18/x2/bn_bias_0"], weights["conv4_18/x2/bn_scale_0"], weights["conv4_18/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_18/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_541 = Conv(flipkernel(weights["conv4_18/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_542 = BatchNorm(identity, weights["conv4_18/x1/bn_bias_0"], weights["conv4_18/x1/bn_scale_0"], weights["conv4_18/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_18/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_543 = (Flux.unsqueeze)(edge_147, (ndims(edge_147) + 1) - 2)
c_544 = (Flux.unsqueeze)(edge_148, (ndims(edge_148) + 1) - 2)
c_545 = (Flux.unsqueeze)(edge_149, (ndims(edge_149) + 1) - 2)
c_546 = (Flux.unsqueeze)(edge_150, (ndims(edge_150) + 1) - 2)
c_547 = Conv(flipkernel(weights["conv4_19/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_548 = BatchNorm(identity, weights["conv4_19/x2/bn_bias_0"], weights["conv4_19/x2/bn_scale_0"], weights["conv4_19/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_19/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_549 = Conv(flipkernel(weights["conv4_19/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_550 = BatchNorm(identity, weights["conv4_19/x1/bn_bias_0"], weights["conv4_19/x1/bn_scale_0"], weights["conv4_19/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_19/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_551 = (Flux.unsqueeze)(edge_151, (ndims(edge_151) + 1) - 2)
c_552 = (Flux.unsqueeze)(edge_152, (ndims(edge_152) + 1) - 2)
c_553 = (Flux.unsqueeze)(edge_153, (ndims(edge_153) + 1) - 2)
c_554 = (Flux.unsqueeze)(edge_154, (ndims(edge_154) + 1) - 2)
c_555 = Conv(flipkernel(weights["conv4_20/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_556 = BatchNorm(identity, weights["conv4_20/x2/bn_bias_0"], weights["conv4_20/x2/bn_scale_0"], weights["conv4_20/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_20/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_557 = Conv(flipkernel(weights["conv4_20/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_558 = BatchNorm(identity, weights["conv4_20/x1/bn_bias_0"], weights["conv4_20/x1/bn_scale_0"], weights["conv4_20/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_20/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_559 = (Flux.unsqueeze)(edge_155, (ndims(edge_155) + 1) - 2)
c_560 = (Flux.unsqueeze)(edge_156, (ndims(edge_156) + 1) - 2)
c_561 = (Flux.unsqueeze)(edge_157, (ndims(edge_157) + 1) - 2)
c_562 = (Flux.unsqueeze)(edge_158, (ndims(edge_158) + 1) - 2)
c_563 = Conv(flipkernel(weights["conv4_21/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_564 = BatchNorm(identity, weights["conv4_21/x2/bn_bias_0"], weights["conv4_21/x2/bn_scale_0"], weights["conv4_21/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_21/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_565 = Conv(flipkernel(weights["conv4_21/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_566 = BatchNorm(identity, weights["conv4_21/x1/bn_bias_0"], weights["conv4_21/x1/bn_scale_0"], weights["conv4_21/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_21/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_567 = (Flux.unsqueeze)(edge_159, (ndims(edge_159) + 1) - 2)
c_568 = (Flux.unsqueeze)(edge_160, (ndims(edge_160) + 1) - 2)
c_569 = (Flux.unsqueeze)(edge_161, (ndims(edge_161) + 1) - 2)
c_570 = (Flux.unsqueeze)(edge_162, (ndims(edge_162) + 1) - 2)
c_571 = Conv(flipkernel(weights["conv4_22/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_572 = BatchNorm(identity, weights["conv4_22/x2/bn_bias_0"], weights["conv4_22/x2/bn_scale_0"], weights["conv4_22/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_22/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_573 = Conv(flipkernel(weights["conv4_22/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_574 = BatchNorm(identity, weights["conv4_22/x1/bn_bias_0"], weights["conv4_22/x1/bn_scale_0"], weights["conv4_22/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_22/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_575 = (Flux.unsqueeze)(edge_163, (ndims(edge_163) + 1) - 2)
c_576 = (Flux.unsqueeze)(edge_164, (ndims(edge_164) + 1) - 2)
c_577 = (Flux.unsqueeze)(edge_165, (ndims(edge_165) + 1) - 2)
c_578 = (Flux.unsqueeze)(edge_166, (ndims(edge_166) + 1) - 2)
c_579 = Conv(flipkernel(weights["conv4_23/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_580 = BatchNorm(identity, weights["conv4_23/x2/bn_bias_0"], weights["conv4_23/x2/bn_scale_0"], weights["conv4_23/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_23/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_581 = Conv(flipkernel(weights["conv4_23/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_582 = BatchNorm(identity, weights["conv4_23/x1/bn_bias_0"], weights["conv4_23/x1/bn_scale_0"], weights["conv4_23/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_23/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_583 = (Flux.unsqueeze)(edge_167, (ndims(edge_167) + 1) - 2)
c_584 = (Flux.unsqueeze)(edge_168, (ndims(edge_168) + 1) - 2)
c_585 = (Flux.unsqueeze)(edge_169, (ndims(edge_169) + 1) - 2)
c_586 = (Flux.unsqueeze)(edge_170, (ndims(edge_170) + 1) - 2)
c_587 = Conv(flipkernel(weights["conv4_24/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_588 = BatchNorm(identity, weights["conv4_24/x2/bn_bias_0"], weights["conv4_24/x2/bn_scale_0"], weights["conv4_24/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_24/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_589 = Conv(flipkernel(weights["conv4_24/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_590 = BatchNorm(identity, weights["conv4_24/x1/bn_bias_0"], weights["conv4_24/x1/bn_scale_0"], weights["conv4_24/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv4_24/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_591 = (Flux.unsqueeze)(edge_171, (ndims(edge_171) + 1) - 2)
c_592 = (Flux.unsqueeze)(edge_172, (ndims(edge_172) + 1) - 2)
c_593 = (Flux.unsqueeze)(edge_173, (ndims(edge_173) + 1) - 2)
c_594 = (Flux.unsqueeze)(edge_174, (ndims(edge_174) + 1) - 2)
c_595 = (Flux.unsqueeze)(edge_175, (ndims(edge_175) + 1) - 2)
c_596 = (Flux.unsqueeze)(edge_176, (ndims(edge_176) + 1) - 2)
c_597 = Conv(flipkernel(weights["conv5_1/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_598 = BatchNorm(identity, weights["conv5_1/x2/bn_bias_0"], weights["conv5_1/x2/bn_scale_0"], weights["conv5_1/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_1/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_599 = Conv(flipkernel(weights["conv5_1/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_600 = BatchNorm(identity, weights["conv5_1/x1/bn_bias_0"], weights["conv5_1/x1/bn_scale_0"], weights["conv5_1/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_1/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_601 = (Flux.unsqueeze)(edge_177, (ndims(edge_177) + 1) - 2)
c_602 = (Flux.unsqueeze)(edge_178, (ndims(edge_178) + 1) - 2)
c_603 = (Flux.unsqueeze)(edge_179, (ndims(edge_179) + 1) - 2)
c_604 = (Flux.unsqueeze)(edge_180, (ndims(edge_180) + 1) - 2)
c_605 = Conv(flipkernel(weights["conv5_2/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_606 = BatchNorm(identity, weights["conv5_2/x2/bn_bias_0"], weights["conv5_2/x2/bn_scale_0"], weights["conv5_2/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_2/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_607 = Conv(flipkernel(weights["conv5_2/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_608 = BatchNorm(identity, weights["conv5_2/x1/bn_bias_0"], weights["conv5_2/x1/bn_scale_0"], weights["conv5_2/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_2/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_609 = (Flux.unsqueeze)(edge_181, (ndims(edge_181) + 1) - 2)
c_610 = (Flux.unsqueeze)(edge_182, (ndims(edge_182) + 1) - 2)
c_611 = (Flux.unsqueeze)(edge_183, (ndims(edge_183) + 1) - 2)
c_612 = (Flux.unsqueeze)(edge_184, (ndims(edge_184) + 1) - 2)
c_613 = Conv(flipkernel(weights["conv5_3/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_614 = BatchNorm(identity, weights["conv5_3/x2/bn_bias_0"], weights["conv5_3/x2/bn_scale_0"], weights["conv5_3/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_3/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_615 = Conv(flipkernel(weights["conv5_3/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_616 = BatchNorm(identity, weights["conv5_3/x1/bn_bias_0"], weights["conv5_3/x1/bn_scale_0"], weights["conv5_3/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_3/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_617 = (Flux.unsqueeze)(edge_185, (ndims(edge_185) + 1) - 2)
c_618 = (Flux.unsqueeze)(edge_186, (ndims(edge_186) + 1) - 2)
c_619 = (Flux.unsqueeze)(edge_187, (ndims(edge_187) + 1) - 2)
c_620 = (Flux.unsqueeze)(edge_188, (ndims(edge_188) + 1) - 2)
c_621 = Conv(flipkernel(weights["conv5_4/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_622 = BatchNorm(identity, weights["conv5_4/x2/bn_bias_0"], weights["conv5_4/x2/bn_scale_0"], weights["conv5_4/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_4/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_623 = Conv(flipkernel(weights["conv5_4/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_624 = BatchNorm(identity, weights["conv5_4/x1/bn_bias_0"], weights["conv5_4/x1/bn_scale_0"], weights["conv5_4/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_4/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_625 = (Flux.unsqueeze)(edge_189, (ndims(edge_189) + 1) - 2)
c_626 = (Flux.unsqueeze)(edge_190, (ndims(edge_190) + 1) - 2)
c_627 = (Flux.unsqueeze)(edge_191, (ndims(edge_191) + 1) - 2)
c_628 = (Flux.unsqueeze)(edge_192, (ndims(edge_192) + 1) - 2)
c_629 = Conv(flipkernel(weights["conv5_5/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_630 = BatchNorm(identity, weights["conv5_5/x2/bn_bias_0"], weights["conv5_5/x2/bn_scale_0"], weights["conv5_5/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_5/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_631 = Conv(flipkernel(weights["conv5_5/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_632 = BatchNorm(identity, weights["conv5_5/x1/bn_bias_0"], weights["conv5_5/x1/bn_scale_0"], weights["conv5_5/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_5/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_633 = (Flux.unsqueeze)(edge_193, (ndims(edge_193) + 1) - 2)
c_634 = (Flux.unsqueeze)(edge_194, (ndims(edge_194) + 1) - 2)
c_635 = (Flux.unsqueeze)(edge_195, (ndims(edge_195) + 1) - 2)
c_636 = (Flux.unsqueeze)(edge_196, (ndims(edge_196) + 1) - 2)
c_637 = Conv(flipkernel(weights["conv5_6/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_638 = BatchNorm(identity, weights["conv5_6/x2/bn_bias_0"], weights["conv5_6/x2/bn_scale_0"], weights["conv5_6/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_6/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_639 = Conv(flipkernel(weights["conv5_6/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_640 = BatchNorm(identity, weights["conv5_6/x1/bn_bias_0"], weights["conv5_6/x1/bn_scale_0"], weights["conv5_6/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_6/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_641 = (Flux.unsqueeze)(edge_197, (ndims(edge_197) + 1) - 2)
c_642 = (Flux.unsqueeze)(edge_198, (ndims(edge_198) + 1) - 2)
c_643 = (Flux.unsqueeze)(edge_199, (ndims(edge_199) + 1) - 2)
c_644 = (Flux.unsqueeze)(edge_200, (ndims(edge_200) + 1) - 2)
c_645 = Conv(flipkernel(weights["conv5_7/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_646 = BatchNorm(identity, weights["conv5_7/x2/bn_bias_0"], weights["conv5_7/x2/bn_scale_0"], weights["conv5_7/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_7/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_647 = Conv(flipkernel(weights["conv5_7/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_648 = BatchNorm(identity, weights["conv5_7/x1/bn_bias_0"], weights["conv5_7/x1/bn_scale_0"], weights["conv5_7/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_7/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_649 = (Flux.unsqueeze)(edge_201, (ndims(edge_201) + 1) - 2)
c_650 = (Flux.unsqueeze)(edge_202, (ndims(edge_202) + 1) - 2)
c_651 = (Flux.unsqueeze)(edge_203, (ndims(edge_203) + 1) - 2)
c_652 = (Flux.unsqueeze)(edge_204, (ndims(edge_204) + 1) - 2)
c_653 = Conv(flipkernel(weights["conv5_8/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_654 = BatchNorm(identity, weights["conv5_8/x2/bn_bias_0"], weights["conv5_8/x2/bn_scale_0"], weights["conv5_8/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_8/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_655 = Conv(flipkernel(weights["conv5_8/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_656 = BatchNorm(identity, weights["conv5_8/x1/bn_bias_0"], weights["conv5_8/x1/bn_scale_0"], weights["conv5_8/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_8/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_657 = (Flux.unsqueeze)(edge_205, (ndims(edge_205) + 1) - 2)
c_658 = (Flux.unsqueeze)(edge_206, (ndims(edge_206) + 1) - 2)
c_659 = (Flux.unsqueeze)(edge_207, (ndims(edge_207) + 1) - 2)
c_660 = (Flux.unsqueeze)(edge_208, (ndims(edge_208) + 1) - 2)
c_661 = Conv(flipkernel(weights["conv5_9/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_662 = BatchNorm(identity, weights["conv5_9/x2/bn_bias_0"], weights["conv5_9/x2/bn_scale_0"], weights["conv5_9/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_9/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_663 = Conv(flipkernel(weights["conv5_9/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_664 = BatchNorm(identity, weights["conv5_9/x1/bn_bias_0"], weights["conv5_9/x1/bn_scale_0"], weights["conv5_9/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_9/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_665 = (Flux.unsqueeze)(edge_209, (ndims(edge_209) + 1) - 2)
c_666 = (Flux.unsqueeze)(edge_210, (ndims(edge_210) + 1) - 2)
c_667 = (Flux.unsqueeze)(edge_211, (ndims(edge_211) + 1) - 2)
c_668 = (Flux.unsqueeze)(edge_212, (ndims(edge_212) + 1) - 2)
c_669 = Conv(flipkernel(weights["conv5_10/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_670 = BatchNorm(identity, weights["conv5_10/x2/bn_bias_0"], weights["conv5_10/x2/bn_scale_0"], weights["conv5_10/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_10/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_671 = Conv(flipkernel(weights["conv5_10/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_672 = BatchNorm(identity, weights["conv5_10/x1/bn_bias_0"], weights["conv5_10/x1/bn_scale_0"], weights["conv5_10/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_10/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_673 = (Flux.unsqueeze)(edge_213, (ndims(edge_213) + 1) - 2)
c_674 = (Flux.unsqueeze)(edge_214, (ndims(edge_214) + 1) - 2)
c_675 = (Flux.unsqueeze)(edge_215, (ndims(edge_215) + 1) - 2)
c_676 = (Flux.unsqueeze)(edge_216, (ndims(edge_216) + 1) - 2)
c_677 = Conv(flipkernel(weights["conv5_11/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_678 = BatchNorm(identity, weights["conv5_11/x2/bn_bias_0"], weights["conv5_11/x2/bn_scale_0"], weights["conv5_11/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_11/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_679 = Conv(flipkernel(weights["conv5_11/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_680 = BatchNorm(identity, weights["conv5_11/x1/bn_bias_0"], weights["conv5_11/x1/bn_scale_0"], weights["conv5_11/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_11/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_681 = (Flux.unsqueeze)(edge_217, (ndims(edge_217) + 1) - 2)
c_682 = (Flux.unsqueeze)(edge_218, (ndims(edge_218) + 1) - 2)
c_683 = (Flux.unsqueeze)(edge_219, (ndims(edge_219) + 1) - 2)
c_684 = (Flux.unsqueeze)(edge_220, (ndims(edge_220) + 1) - 2)
c_685 = Conv(flipkernel(weights["conv5_12/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_686 = BatchNorm(identity, weights["conv5_12/x2/bn_bias_0"], weights["conv5_12/x2/bn_scale_0"], weights["conv5_12/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_12/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_687 = Conv(flipkernel(weights["conv5_12/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_688 = BatchNorm(identity, weights["conv5_12/x1/bn_bias_0"], weights["conv5_12/x1/bn_scale_0"], weights["conv5_12/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_12/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_689 = (Flux.unsqueeze)(edge_221, (ndims(edge_221) + 1) - 2)
c_690 = (Flux.unsqueeze)(edge_222, (ndims(edge_222) + 1) - 2)
c_691 = (Flux.unsqueeze)(edge_223, (ndims(edge_223) + 1) - 2)
c_692 = (Flux.unsqueeze)(edge_224, (ndims(edge_224) + 1) - 2)
c_693 = Conv(flipkernel(weights["conv5_13/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_694 = BatchNorm(identity, weights["conv5_13/x2/bn_bias_0"], weights["conv5_13/x2/bn_scale_0"], weights["conv5_13/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_13/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_695 = Conv(flipkernel(weights["conv5_13/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_696 = BatchNorm(identity, weights["conv5_13/x1/bn_bias_0"], weights["conv5_13/x1/bn_scale_0"], weights["conv5_13/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_13/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_697 = (Flux.unsqueeze)(edge_225, (ndims(edge_225) + 1) - 2)
c_698 = (Flux.unsqueeze)(edge_226, (ndims(edge_226) + 1) - 2)
c_699 = (Flux.unsqueeze)(edge_227, (ndims(edge_227) + 1) - 2)
c_700 = (Flux.unsqueeze)(edge_228, (ndims(edge_228) + 1) - 2)
c_701 = Conv(flipkernel(weights["conv5_14/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_702 = BatchNorm(identity, weights["conv5_14/x2/bn_bias_0"], weights["conv5_14/x2/bn_scale_0"], weights["conv5_14/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_14/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_703 = Conv(flipkernel(weights["conv5_14/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_704 = BatchNorm(identity, weights["conv5_14/x1/bn_bias_0"], weights["conv5_14/x1/bn_scale_0"], weights["conv5_14/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_14/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_705 = (Flux.unsqueeze)(edge_229, (ndims(edge_229) + 1) - 2)
c_706 = (Flux.unsqueeze)(edge_230, (ndims(edge_230) + 1) - 2)
c_707 = (Flux.unsqueeze)(edge_231, (ndims(edge_231) + 1) - 2)
c_708 = (Flux.unsqueeze)(edge_232, (ndims(edge_232) + 1) - 2)
c_709 = Conv(flipkernel(weights["conv5_15/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_710 = BatchNorm(identity, weights["conv5_15/x2/bn_bias_0"], weights["conv5_15/x2/bn_scale_0"], weights["conv5_15/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_15/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_711 = Conv(flipkernel(weights["conv5_15/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_712 = BatchNorm(identity, weights["conv5_15/x1/bn_bias_0"], weights["conv5_15/x1/bn_scale_0"], weights["conv5_15/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_15/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_713 = (Flux.unsqueeze)(edge_233, (ndims(edge_233) + 1) - 2)
c_714 = (Flux.unsqueeze)(edge_234, (ndims(edge_234) + 1) - 2)
c_715 = (Flux.unsqueeze)(edge_235, (ndims(edge_235) + 1) - 2)
c_716 = (Flux.unsqueeze)(edge_236, (ndims(edge_236) + 1) - 2)
c_717 = Conv(flipkernel(weights["conv5_16/x2_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(1, 1), dilation=(1, 1))
c_718 = BatchNorm(identity, weights["conv5_16/x2/bn_bias_0"], weights["conv5_16/x2/bn_scale_0"], weights["conv5_16/x2/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_16/x2/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_719 = Conv(flipkernel(weights["conv5_16/x1_w_0"]), Float32[0.0], relu, stride=(1, 1), pad=(0, 0), dilation=(1, 1))
c_720 = BatchNorm(identity, weights["conv5_16/x1/bn_bias_0"], weights["conv5_16/x1/bn_scale_0"], weights["conv5_16/x1/bn_mean_0"], broadcast(Float32, broadcast(sqrt, broadcast(+, 1.0f-5, weights["conv5_16/x1/bn_var_0"]))), 9.999999747378752e-6, 0.9, false)
c_721 = (Flux.unsqueeze)(edge_237, (ndims(edge_237) + 1) - 2)
c_722 = (Flux.unsqueeze)(edge_238, (ndims(edge_238) + 1) - 2)
c_723 = (Flux.unsqueeze)(edge_239, (ndims(edge_239) + 1) - 2)
c_724 = (Flux.unsqueeze)(edge_240, (ndims(edge_240) + 1) - 2)
c_725 = (Flux.unsqueeze)(edge_241, (ndims(edge_241) + 1) - 2)
c_726 = (Flux.unsqueeze)(edge_242, (ndims(edge_242) + 1) - 2)
(x_727,)->begin
edge_728 = maxpool(relu.(c_251(c_252(x_727)) .* c_253 .+ c_254), (3, 3), pad=(1, 1), stride=(2, 2))
edge_729 = cat(3, edge_728, c_255(relu.(c_256(c_257(relu.(c_258(edge_728) .* c_259 .+ c_260))) .* c_261 .+ c_262)))
edge_730 = cat(3, edge_729, c_263(relu.(c_264(c_265(relu.(c_266(edge_729) .* c_267 .+ c_268))) .* c_269 .+ c_270)))
edge_731 = cat(3, edge_730, c_271(relu.(c_272(c_273(relu.(c_274(edge_730) .* c_275 .+ c_276))) .* c_277 .+ c_278)))
edge_732 = cat(3, edge_731, c_279(relu.(c_280(c_281(relu.(c_282(edge_731) .* c_283 .+ c_284))) .* c_285 .+ c_286)))
edge_733 = cat(3, edge_732, c_287(relu.(c_288(c_289(relu.(c_290(edge_732) .* c_291 .+ c_292))) .* c_293 .+ c_294)))
edge_734 = meanpool(c_249(relu.(c_250(cat(3, edge_733, c_295(relu.(c_296(c_297(relu.(c_298(edge_733) .* c_299 .+ c_300))) .* c_301 .+ c_302)))) .* c_303 .+ c_304)), (2, 2), pad=(0, 0), stride=(2, 2))
edge_735 = cat(3, edge_734, c_305(relu.(c_306(c_307(relu.(c_308(edge_734) .* c_309 .+ c_310))) .* c_311 .+ c_312)))
edge_736 = cat(3, edge_735, c_313(relu.(c_314(c_315(relu.(c_316(edge_735) .* c_317 .+ c_318))) .* c_319 .+ c_320)))
edge_737 = cat(3, edge_736, c_321(relu.(c_322(c_323(relu.(c_324(edge_736) .* c_325 .+ c_326))) .* c_327 .+ c_328)))
edge_738 = cat(3, edge_737, c_329(relu.(c_330(c_331(relu.(c_332(edge_737) .* c_333 .+ c_334))) .* c_335 .+ c_336)))
edge_739 = cat(3, edge_738, c_337(relu.(c_338(c_339(relu.(c_340(edge_738) .* c_341 .+ c_342))) .* c_343 .+ c_344)))
edge_740 = cat(3, edge_739, c_345(relu.(c_346(c_347(relu.(c_348(edge_739) .* c_349 .+ c_350))) .* c_351 .+ c_352)))
edge_741 = cat(3, edge_740, c_353(relu.(c_354(c_355(relu.(c_356(edge_740) .* c_357 .+ c_358))) .* c_359 .+ c_360)))
edge_742 = cat(3, edge_741, c_361(relu.(c_362(c_363(relu.(c_364(edge_741) .* c_365 .+ c_366))) .* c_367 .+ c_368)))
edge_743 = cat(3, edge_742, c_369(relu.(c_370(c_371(relu.(c_372(edge_742) .* c_373 .+ c_374))) .* c_375 .+ c_376)))
edge_744 = cat(3, edge_743, c_377(relu.(c_378(c_379(relu.(c_380(edge_743) .* c_381 .+ c_382))) .* c_383 .+ c_384)))
edge_745 = cat(3, edge_744, c_385(relu.(c_386(c_387(relu.(c_388(edge_744) .* c_389 .+ c_390))) .* c_391 .+ c_392)))
edge_746 = meanpool(c_247(relu.(c_248(cat(3, edge_745, c_393(relu.(c_394(c_395(relu.(c_396(edge_745) .* c_397 .+ c_398))) .* c_399 .+ c_400)))) .* c_401 .+ c_402)), (2, 2), pad=(0, 0), stride=(2, 2))
edge_747 = cat(3, edge_746, c_403(relu.(c_404(c_405(relu.(c_406(edge_746) .* c_407 .+ c_408))) .* c_409 .+ c_410)))
edge_748 = cat(3, edge_747, c_411(relu.(c_412(c_413(relu.(c_414(edge_747) .* c_415 .+ c_416))) .* c_417 .+ c_418)))
edge_749 = cat(3, edge_748, c_419(relu.(c_420(c_421(relu.(c_422(edge_748) .* c_423 .+ c_424))) .* c_425 .+ c_426)))
edge_750 = cat(3, edge_749, c_427(relu.(c_428(c_429(relu.(c_430(edge_749) .* c_431 .+ c_432))) .* c_433 .+ c_434)))
edge_751 = cat(3, edge_750, c_435(relu.(c_436(c_437(relu.(c_438(edge_750) .* c_439 .+ c_440))) .* c_441 .+ c_442)))
edge_752 = cat(3, edge_751, c_443(relu.(c_444(c_445(relu.(c_446(edge_751) .* c_447 .+ c_448))) .* c_449 .+ c_450)))
edge_753 = cat(3, edge_752, c_451(relu.(c_452(c_453(relu.(c_454(edge_752) .* c_455 .+ c_456))) .* c_457 .+ c_458)))
edge_754 = cat(3, edge_753, c_459(relu.(c_460(c_461(relu.(c_462(edge_753) .* c_463 .+ c_464))) .* c_465 .+ c_466)))
edge_755 = cat(3, edge_754, c_467(relu.(c_468(c_469(relu.(c_470(edge_754) .* c_471 .+ c_472))) .* c_473 .+ c_474)))
edge_756 = cat(3, edge_755, c_475(relu.(c_476(c_477(relu.(c_478(edge_755) .* c_479 .+ c_480))) .* c_481 .+ c_482)))
edge_757 = cat(3, edge_756, c_483(relu.(c_484(c_485(relu.(c_486(edge_756) .* c_487 .+ c_488))) .* c_489 .+ c_490)))
edge_758 = cat(3, edge_757, c_491(relu.(c_492(c_493(relu.(c_494(edge_757) .* c_495 .+ c_496))) .* c_497 .+ c_498)))
edge_759 = cat(3, edge_758, c_499(relu.(c_500(c_501(relu.(c_502(edge_758) .* c_503 .+ c_504))) .* c_505 .+ c_506)))
edge_760 = cat(3, edge_759, c_507(relu.(c_508(c_509(relu.(c_510(edge_759) .* c_511 .+ c_512))) .* c_513 .+ c_514)))
edge_761 = cat(3, edge_760, c_515(relu.(c_516(c_517(relu.(c_518(edge_760) .* c_519 .+ c_520))) .* c_521 .+ c_522)))
edge_762 = cat(3, edge_761, c_523(relu.(c_524(c_525(relu.(c_526(edge_761) .* c_527 .+ c_528))) .* c_529 .+ c_530)))
edge_763 = cat(3, edge_762, c_531(relu.(c_532(c_533(relu.(c_534(edge_762) .* c_535 .+ c_536))) .* c_537 .+ c_538)))
edge_764 = cat(3, edge_763, c_539(relu.(c_540(c_541(relu.(c_542(edge_763) .* c_543 .+ c_544))) .* c_545 .+ c_546)))
edge_765 = cat(3, edge_764, c_547(relu.(c_548(c_549(relu.(c_550(edge_764) .* c_551 .+ c_552))) .* c_553 .+ c_554)))
edge_766 = cat(3, edge_765, c_555(relu.(c_556(c_557(relu.(c_558(edge_765) .* c_559 .+ c_560))) .* c_561 .+ c_562)))
edge_767 = cat(3, edge_766, c_563(relu.(c_564(c_565(relu.(c_566(edge_766) .* c_567 .+ c_568))) .* c_569 .+ c_570)))
edge_768 = cat(3, edge_767, c_571(relu.(c_572(c_573(relu.(c_574(edge_767) .* c_575 .+ c_576))) .* c_577 .+ c_578)))
edge_769 = cat(3, edge_768, c_579(relu.(c_580(c_581(relu.(c_582(edge_768) .* c_583 .+ c_584))) .* c_585 .+ c_586)))
edge_770 = meanpool(c_245(relu.(c_246(cat(3, edge_769, c_587(relu.(c_588(c_589(relu.(c_590(edge_769) .* c_591 .+ c_592))) .* c_593 .+ c_594)))) .* c_595 .+ c_596)), (2, 2), pad=(0, 0), stride=(2, 2))
edge_771 = cat(3, edge_770, c_597(relu.(c_598(c_599(relu.(c_600(edge_770) .* c_601 .+ c_602))) .* c_603 .+ c_604)))
edge_772 = cat(3, edge_771, c_605(relu.(c_606(c_607(relu.(c_608(edge_771) .* c_609 .+ c_610))) .* c_611 .+ c_612)))
edge_773 = cat(3, edge_772, c_613(relu.(c_614(c_615(relu.(c_616(edge_772) .* c_617 .+ c_618))) .* c_619 .+ c_620)))
edge_774 = cat(3, edge_773, c_621(relu.(c_622(c_623(relu.(c_624(edge_773) .* c_625 .+ c_626))) .* c_627 .+ c_628)))
edge_775 = cat(3, edge_774, c_629(relu.(c_630(c_631(relu.(c_632(edge_774) .* c_633 .+ c_634))) .* c_635 .+ c_636)))
edge_776 = cat(3, edge_775, c_637(relu.(c_638(c_639(relu.(c_640(edge_775) .* c_641 .+ c_642))) .* c_643 .+ c_644)))
edge_777 = cat(3, edge_776, c_645(relu.(c_646(c_647(relu.(c_648(edge_776) .* c_649 .+ c_650))) .* c_651 .+ c_652)))
edge_778 = cat(3, edge_777, c_653(relu.(c_654(c_655(relu.(c_656(edge_777) .* c_657 .+ c_658))) .* c_659 .+ c_660)))
edge_779 = cat(3, edge_778, c_661(relu.(c_662(c_663(relu.(c_664(edge_778) .* c_665 .+ c_666))) .* c_667 .+ c_668)))
edge_780 = cat(3, edge_779, c_669(relu.(c_670(c_671(relu.(c_672(edge_779) .* c_673 .+ c_674))) .* c_675 .+ c_676)))
edge_781 = cat(3, edge_780, c_677(relu.(c_678(c_679(relu.(c_680(edge_780) .* c_681 .+ c_682))) .* c_683 .+ c_684)))
edge_782 = cat(3, edge_781, c_685(relu.(c_686(c_687(relu.(c_688(edge_781) .* c_689 .+ c_690))) .* c_691 .+ c_692)))
edge_783 = cat(3, edge_782, c_693(relu.(c_694(c_695(relu.(c_696(edge_782) .* c_697 .+ c_698))) .* c_699 .+ c_700)))
edge_784 = cat(3, edge_783, c_701(relu.(c_702(c_703(relu.(c_704(edge_783) .* c_705 .+ c_706))) .* c_707 .+ c_708)))
edge_785 = cat(3, edge_784, c_709(relu.(c_710(c_711(relu.(c_712(edge_784) .* c_713 .+ c_714))) .* c_715 .+ c_716)))
c_243(mean(relu.(c_244(cat(3, edge_785, c_717(relu.(c_718(c_719(relu.(c_720(edge_785) .* c_721 .+ c_722))) .* c_723 .+ c_724)))) .* c_725 .+ c_726), (1, 2)))
end
end
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment