segmentation_1_4_0_fp32_combined/vaiml_par_0/OnnxModel.onnx.mlir

#loc = loc(unknown)
module attributes {
  llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128",
  llvm.target_triple = "x86_64-unknown-linux-gnu",
  "onnx-mlir.symbol-postfix" = "onnxmodel.onnx.mlir",
  vaimlconf.device = "stx",
  vaimlconf.device_models = "${vaimlconf.install_dir}/data/deviceModels",
  vaimlconf.install_dir = "/usr/local/lib/python3.10/dist-packages/flexml/flexml_extras",
  vaimlconf.library_metadata = ["${vaimlconf.install_dir}/data/libraryMetadata/L1", "${vaimlconf.install_dir}/data/libraryMetadata/L2", "${vaimlconf.install_dir}/../../vitis_mllib/L1/metadata", "${vaimlconf.install_dir}/../../vitis_mllib/L2/metadata", "${vaimlconf.install_dir}/share/microkernel-tiling/tiling-recipe-specs"],
  vaimlconf.single_core_compiler = "chess"} {
  func.func @main_graph(%arg0: tensor<1x180x320x4xf32> {onnx.name = "385"} loc(unknown), %arg1: tensor<1x16x90x160xf32> {onnx.name = "394"} loc(unknown), %arg2: tensor<1x20x45x80xf32> {onnx.name = "395"} loc(unknown), %arg3: tensor<1x40x23x40xf32> {onnx.name = "396"} loc(unknown), %arg4: tensor<1x64x12x20xf32> {onnx.name = "397"} loc(unknown)) -> (tensor<1x1x180x320xf32> {onnx.name = "921"}, tensor<1x16x90x160xf32> {onnx.name = "894"}, tensor<1x20x45x80xf32> {onnx.name = "868"}, tensor<1x40x23x40xf32> {onnx.name = "832"}, tensor<1x64x12x20xf32> {onnx.name = "796"}, tensor<1x3x180x320xf32> {onnx.name = "916"}) {
    %0 = onnx.Constant dense<[[[-4.850000e-01]], [[-4.560000e-01]], [[-4.060000e-01]]]> : tensor<3x1x1xf32> loc(#loc)
    %1 = onnx.Constant dense<[3, 1]> : tensor<2xi64> loc(#loc)
    %2 = onnx.Constant dense<16> : tensor<2xi64> loc(#loc)
    %3 = onnx.Constant dense<20> : tensor<2xi64> loc(#loc)
    %4 = onnx.Constant dense<40> : tensor<2xi64> loc(#loc)
    %5 = onnx.Constant dense<64> : tensor<2xi64> loc(#loc)
    %6 = "onnx.NoValue"() {onnx_node_name = "onnx.NoValue_33", value} : () -> none loc(#loc)
    %7 = onnx.Constant dense_resource<__elided__> : tensor<184x80x1x1xf32> loc(#loc1)
    %8 = onnx.Constant dense_resource<__elided__> : tensor<184xf32> loc(#loc2)
    %9 = onnx.Constant dense_resource<__elided__> : tensor<184x1x3x3xf32> loc(#loc3)
    %10 = onnx.Constant dense_resource<__elided__> : tensor<184xf32> loc(#loc4)
    %11 = onnx.Constant dense_resource<__elided__> : tensor<80x184x1x1xf32> loc(#loc5)
    %12 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc6)
    %13 = onnx.Constant dense_resource<__elided__> : tensor<184x80x1x1xf32> loc(#loc7)
    %14 = onnx.Constant dense_resource<__elided__> : tensor<184xf32> loc(#loc8)
    %15 = onnx.Constant dense_resource<__elided__> : tensor<184x1x3x3xf32> loc(#loc9)
    %16 = onnx.Constant dense_resource<__elided__> : tensor<184xf32> loc(#loc10)
    %17 = onnx.Constant dense_resource<__elided__> : tensor<80x184x1x1xf32> loc(#loc11)
    %18 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc12)
    %19 = onnx.Constant dense_resource<__elided__> : tensor<480x80x1x1xf32> loc(#loc13)
    %20 = onnx.Constant dense_resource<__elided__> : tensor<480xf32> loc(#loc14)
    %21 = onnx.Constant dense_resource<__elided__> : tensor<480x1x3x3xf32> loc(#loc15)
    %22 = onnx.Constant dense_resource<__elided__> : tensor<480xf32> loc(#loc16)
    %23 = onnx.Constant dense_resource<__elided__> : tensor<112x480x1x1xf32> loc(#loc17)
    %24 = onnx.Constant dense_resource<__elided__> : tensor<112xf32> loc(#loc18)
    %25 = onnx.Constant dense_resource<__elided__> : tensor<672x112x1x1xf32> loc(#loc19)
    %26 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc20)
    %27 = onnx.Constant dense_resource<__elided__> : tensor<672x1x3x3xf32> loc(#loc21)
    %28 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc22)
    %29 = onnx.Constant dense_resource<__elided__> : tensor<112x672x1x1xf32> loc(#loc23)
    %30 = onnx.Constant dense_resource<__elided__> : tensor<112xf32> loc(#loc24)
    %31 = onnx.Constant dense_resource<__elided__> : tensor<672x112x1x1xf32> loc(#loc25)
    %32 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc26)
    %33 = onnx.Constant dense_resource<__elided__> : tensor<672x1x5x5xf32> loc(#loc27)
    %34 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc28)
    %35 = onnx.Constant dense_resource<__elided__> : tensor<160x672x1x1xf32> loc(#loc29)
    %36 = onnx.Constant dense_resource<__elided__> : tensor<160xf32> loc(#loc30)
    %37 = onnx.Constant dense_resource<__elided__> : tensor<960x160x1x1xf32> loc(#loc31)
    %38 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc32)
    %39 = onnx.Constant dense_resource<__elided__> : tensor<960x1x5x5xf32> loc(#loc33)
    %40 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc34)
    %41 = onnx.Constant dense_resource<__elided__> : tensor<160x960x1x1xf32> loc(#loc35)
    %42 = onnx.Constant dense_resource<__elided__> : tensor<160xf32> loc(#loc36)
    %43 = onnx.Constant dense_resource<__elided__> : tensor<960x160x1x1xf32> loc(#loc37)
    %44 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc38)
    %45 = onnx.Constant dense_resource<__elided__> : tensor<960x1x5x5xf32> loc(#loc39)
    %46 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc40)
    %47 = onnx.Constant dense_resource<__elided__> : tensor<160x960x1x1xf32> loc(#loc41)
    %48 = onnx.Constant dense_resource<__elided__> : tensor<160xf32> loc(#loc42)
    %49 = onnx.Constant dense_resource<__elided__> : tensor<960x160x1x1xf32> loc(#loc43)
    %50 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc44)
    %51 = onnx.Constant dense_resource<__elided__> : tensor<128x960x1x1xf32> loc(#loc45)
    %52 = onnx.Constant dense_resource<__elided__> : tensor<128xf32> loc(#loc46)
    %53 = onnx.Constant dense_resource<__elided__> : tensor<80x171x3x3xf32> loc(#loc47)
    %54 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc48)
    %55 = onnx.Constant dense_resource<__elided__> : tensor<40x107x3x3xf32> loc(#loc49)
    %56 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc50)
    %57 = onnx.Constant dense_resource<__elided__> : tensor<32x59x3x3xf32> loc(#loc51)
    %58 = onnx.Constant dense_resource<__elided__> : tensor<32xf32> loc(#loc52)
    %59 = onnx.Constant dense_resource<__elided__> : tensor<16x35x3x3xf32> loc(#loc53)
    %60 = onnx.Constant dense<[0.281471759, -0.0896756947, 0.0517602414, -0.266139954, 0.132527292, 0.684469878, -0.0511226803, 0.859402895, 0.504835129, 0.569725394, 0.217058718, -0.0543790609, -0.30986914, 0.451566547, 0.166573063, 0.415171683]> : tensor<16xf32> loc(#loc54)
    %61 = onnx.Constant dense_resource<__elided__> : tensor<16x16x3x3xf32> loc(#loc55)
    %62 = onnx.Constant dense<[-0.257117867, -0.332320929, -0.342930794, 0.882337093, -0.811691761, 1.04650748, 1.993430e-01, 0.471133053, 0.0722430944, 0.554342687, 1.3374486, 0.48697716, 1.31853354, 0.714223623, 1.16618729, 0.738572299]> : tensor<16xf32> loc(#loc56)
    %63 = onnx.Constant dense<2> : tensor<1xi64> loc(#loc57)
    %64 = onnx.Constant dense<[1.000000e+00, 1.000000e+00, 2.000000e+00, 2.000000e+00]> : tensor<4xf32> loc(#loc58)
    %65 = onnx.Constant dense<2.550000e+02> : tensor<f32> loc(#loc59)
    %66 = onnx.Constant dense<3> : tensor<1xi64> loc(#loc60)
    %67 = onnx.Constant dense<0> : tensor<1xi64> loc(#loc61)
    %68 = onnx.Constant dense<[[[2.290000e-01]], [[2.240000e-01]], [[2.250000e-01]]]> : tensor<3x1x1xf32> loc(#loc62)
    %69 = onnx.Constant dense<0.000000e+00> : tensor<f32> loc(#loc63)
    %70 = onnx.Constant dense<6.000000e+00> : tensor<f32> loc(#loc64)
    %71 = onnx.Constant dense<3.000000e+00> : tensor<f32> loc(#loc65)
    %72 = onnx.Constant dense<23> : tensor<1xi64> loc(#loc66)
    %73 = onnx.Constant dense<45> : tensor<1xi64> loc(#loc67)
    %74 = onnx.Constant dense<1> : tensor<1xi64> loc(#loc68)
    %75 = onnx.Constant dense<1.000000e+00> : tensor<f32> loc(#loc69)
    %76 = onnx.Constant dense_resource<__elided__> : tensor<16x3x3x3xf32> loc(#loc70)
    %77 = onnx.Constant dense<[2.98861408, -1.22985208, 2.43826318, -3.98499513, 4.62797928, 2.54142761, 2.45345306, 2.64061832, 2.13576674, 2.30800247, -0.198341176, -0.427822977, -1.09159482, 4.85548782, 2.70597649, 2.6902504]> : tensor<16xf32> loc(#loc71)
    %78 = onnx.Constant dense_resource<__elided__> : tensor<16x1x3x3xf32> loc(#loc72)
    %79 = onnx.Constant dense<[-4.38406658, -1.06764766E-8, -0.704851329, -1.05036237E-8, -4.89120433E-9, 1.53249037, -0.0617836975, 2.16366434, 0.0416259095, -4.12739087E-9, -3.50249429E-9, -7.75795516E-9, -4.04315559E-9, 0.292217016, -0.010752866, 1.63358212]> : tensor<16xf32> loc(#loc73)
    %80 = onnx.Constant dense_resource<__elided__> : tensor<16x16x1x1xf32> loc(#loc74)
    %81 = onnx.Constant dense<[-1.31068802, 0.586562276, 5.67538071, 0.551027656, 2.19523954, 3.83854461, 0.0600251146, -2.18778157, -1.5404067, 2.044780e+00, -4.23846388, 0.703142225, -8.39978456E-5, 3.50620365, -0.531753063, -5.91183185]> : tensor<16xf32> loc(#loc75)
    %82 = onnx.Constant dense_resource<__elided__> : tensor<64x16x1x1xf32> loc(#loc76)
    %83 = onnx.Constant dense_resource<__elided__> : tensor<64xf32> loc(#loc77)
    %84 = onnx.Constant dense_resource<__elided__> : tensor<64x1x3x3xf32> loc(#loc78)
    %85 = onnx.Constant dense_resource<__elided__> : tensor<64xf32> loc(#loc79)
    %86 = onnx.Constant dense_resource<__elided__> : tensor<24x64x1x1xf32> loc(#loc80)
    %87 = onnx.Constant dense_resource<__elided__> : tensor<24xf32> loc(#loc81)
    %88 = onnx.Constant dense_resource<__elided__> : tensor<72x24x1x1xf32> loc(#loc82)
    %89 = onnx.Constant dense_resource<__elided__> : tensor<72xf32> loc(#loc83)
    %90 = onnx.Constant dense_resource<__elided__> : tensor<72x1x3x3xf32> loc(#loc84)
    %91 = onnx.Constant dense_resource<__elided__> : tensor<72xf32> loc(#loc85)
    %92 = onnx.Constant dense_resource<__elided__> : tensor<24x72x1x1xf32> loc(#loc86)
    %93 = onnx.Constant dense_resource<__elided__> : tensor<24xf32> loc(#loc87)
    %94 = onnx.Constant dense_resource<__elided__> : tensor<72x24x1x1xf32> loc(#loc88)
    %95 = onnx.Constant dense_resource<__elided__> : tensor<72xf32> loc(#loc89)
    %96 = onnx.Constant dense_resource<__elided__> : tensor<72x1x5x5xf32> loc(#loc90)
    %97 = onnx.Constant dense_resource<__elided__> : tensor<72xf32> loc(#loc91)
    %98 = onnx.Constant dense_resource<__elided__> : tensor<40x72x1x1xf32> loc(#loc92)
    %99 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc93)
    %100 = onnx.Constant dense_resource<__elided__> : tensor<120x40x1x1xf32> loc(#loc94)
    %101 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc95)
    %102 = onnx.Constant dense_resource<__elided__> : tensor<120x1x5x5xf32> loc(#loc96)
    %103 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc97)
    %104 = onnx.Constant dense_resource<__elided__> : tensor<40x120x1x1xf32> loc(#loc98)
    %105 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc99)
    %106 = onnx.Constant dense_resource<__elided__> : tensor<120x40x1x1xf32> loc(#loc100)
    %107 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc101)
    %108 = onnx.Constant dense_resource<__elided__> : tensor<120x1x5x5xf32> loc(#loc102)
    %109 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc103)
    %110 = onnx.Constant dense_resource<__elided__> : tensor<40x120x1x1xf32> loc(#loc104)
    %111 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc105)
    %112 = onnx.Constant dense_resource<__elided__> : tensor<240x40x1x1xf32> loc(#loc106)
    %113 = onnx.Constant dense_resource<__elided__> : tensor<240xf32> loc(#loc107)
    %114 = onnx.Constant dense_resource<__elided__> : tensor<240x1x3x3xf32> loc(#loc108)
    %115 = onnx.Constant dense_resource<__elided__> : tensor<240xf32> loc(#loc109)
    %116 = onnx.Constant dense_resource<__elided__> : tensor<80x240x1x1xf32> loc(#loc110)
    %117 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc111)
    %118 = onnx.Constant dense_resource<__elided__> : tensor<200x80x1x1xf32> loc(#loc112)
    %119 = onnx.Constant dense_resource<__elided__> : tensor<200xf32> loc(#loc113)
    %120 = onnx.Constant dense_resource<__elided__> : tensor<200x1x3x3xf32> loc(#loc114)
    %121 = onnx.Constant dense_resource<__elided__> : tensor<200xf32> loc(#loc115)
    %122 = onnx.Constant dense_resource<__elided__> : tensor<80x200x1x1xf32> loc(#loc116)
    %123 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc117)
    %124 = onnx.Constant dense_resource<__elided__> : tensor<128x960x1x1xf32> loc(#loc118)
    %125 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc119)
    %126 = onnx.Constant dense_resource<__elided__> : tensor<120x480x1x1xf32> loc(#loc120)
    %127 = onnx.Constant dense_resource<__elided__> : tensor<480xf32> loc(#loc121)
    %128 = onnx.Constant dense_resource<__elided__> : tensor<480x120x1x1xf32> loc(#loc122)
    %129 = onnx.Constant dense_resource<__elided__> : tensor<168xf32> loc(#loc123)
    %130 = onnx.Constant dense_resource<__elided__> : tensor<168x672x1x1xf32> loc(#loc124)
    %131 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc125)
    %132 = onnx.Constant dense_resource<__elided__> : tensor<672x168x1x1xf32> loc(#loc126)
    %133 = onnx.Constant dense_resource<__elided__> : tensor<168xf32> loc(#loc127)
    %134 = onnx.Constant dense_resource<__elided__> : tensor<168x672x1x1xf32> loc(#loc128)
    %135 = onnx.Constant dense_resource<__elided__> : tensor<672xf32> loc(#loc129)
    %136 = onnx.Constant dense_resource<__elided__> : tensor<672x168x1x1xf32> loc(#loc130)
    %137 = onnx.Constant dense_resource<__elided__> : tensor<240xf32> loc(#loc131)
    %138 = onnx.Constant dense_resource<__elided__> : tensor<240x960x1x1xf32> loc(#loc132)
    %139 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc133)
    %140 = onnx.Constant dense_resource<__elided__> : tensor<960x240x1x1xf32> loc(#loc134)
    %141 = onnx.Constant dense_resource<__elided__> : tensor<240xf32> loc(#loc135)
    %142 = onnx.Constant dense_resource<__elided__> : tensor<240x960x1x1xf32> loc(#loc136)
    %143 = onnx.Constant dense_resource<__elided__> : tensor<960xf32> loc(#loc137)
    %144 = onnx.Constant dense_resource<__elided__> : tensor<960x240x1x1xf32> loc(#loc138)
    %145 = onnx.Constant dense_resource<__elided__> : tensor<24xf32> loc(#loc139)
    %146 = onnx.Constant dense_resource<__elided__> : tensor<24x72x1x1xf32> loc(#loc140)
    %147 = onnx.Constant dense_resource<__elided__> : tensor<72xf32> loc(#loc141)
    %148 = onnx.Constant dense_resource<__elided__> : tensor<72x24x1x1xf32> loc(#loc142)
    %149 = onnx.Constant dense_resource<__elided__> : tensor<32xf32> loc(#loc143)
    %150 = onnx.Constant dense_resource<__elided__> : tensor<32x120x1x1xf32> loc(#loc144)
    %151 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc145)
    %152 = onnx.Constant dense_resource<__elided__> : tensor<120x32x1x1xf32> loc(#loc146)
    %153 = onnx.Constant dense_resource<__elided__> : tensor<32xf32> loc(#loc147)
    %154 = onnx.Constant dense_resource<__elided__> : tensor<32x120x1x1xf32> loc(#loc148)
    %155 = onnx.Constant dense_resource<__elided__> : tensor<120xf32> loc(#loc149)
    %156 = onnx.Constant dense_resource<__elided__> : tensor<120x32x1x1xf32> loc(#loc150)
    %157 = onnx.Constant dense<[0.00544366054, 0.154367775, 0.115729354, 0.171141103, -0.168815523, 0.0456937179, 0.188233331, 0.0151384082, 0.242783383, -0.139173314, -0.24988465, -9.479440e-02, -0.055940561, -0.0512795448, -0.0738077834, 0.0476587117]> : tensor<16xf32> loc(#loc151)
    %158 = onnx.Constant dense_resource<__elided__> : tensor<16x32x3x3xf32> loc(#loc152)
    %159 = onnx.Constant dense_resource<__elided__> : tensor<32xf32> loc(#loc153)
    %160 = onnx.Constant dense_resource<__elided__> : tensor<32x32x3x3xf32> loc(#loc154)
    %161 = onnx.Constant dense_resource<__elided__> : tensor<20xf32> loc(#loc155)
    %162 = onnx.Constant dense_resource<__elided__> : tensor<20x40x3x3xf32> loc(#loc156)
    %163 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc157)
    %164 = onnx.Constant dense_resource<__elided__> : tensor<40x40x3x3xf32> loc(#loc158)
    %165 = onnx.Constant dense_resource<__elided__> : tensor<40xf32> loc(#loc159)
    %166 = onnx.Constant dense_resource<__elided__> : tensor<40x80x3x3xf32> loc(#loc160)
    %167 = onnx.Constant dense_resource<__elided__> : tensor<80xf32> loc(#loc161)
    %168 = onnx.Constant dense_resource<__elided__> : tensor<80x80x3x3xf32> loc(#loc162)
    %169 = onnx.Constant dense_resource<__elided__> : tensor<64xf32> loc(#loc163)
    %170 = onnx.Constant dense_resource<__elided__> : tensor<64x128x3x3xf32> loc(#loc164)
    %171 = onnx.Constant dense_resource<__elided__> : tensor<128xf32> loc(#loc165)
    %172 = onnx.Constant dense_resource<__elided__> : tensor<128x128x3x3xf32> loc(#loc166)
    %173 = onnx.Constant dense<[0.00409470545, 0.00284675183, 0.00200544903, 0.124928087]> : tensor<4xf32> loc(#loc167)
    %174 = onnx.Constant dense_resource<__elided__> : tensor<4x16x1x1xf32> loc(#loc168)
    %175 = "onnx.Div"(%arg0, %65) {onnx_node_name = "Div_2"} : (tensor<1x180x320x4xf32>, tensor<f32>) -> tensor<1x180x320x4xf32> loc(#loc169)
    %176 = "onnx.Slice"(%175, %67, %66, %66, %74) {onnx_node_name = "Slice_7"} : (tensor<1x180x320x4xf32>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>) -> tensor<1x180x320x3xf32> loc(#loc170)
    %177 = "onnx.Transpose"(%176) {onnx_node_name = "Transpose_8", perm = [0, 3, 1, 2]} : (tensor<1x180x320x3xf32>) -> tensor<1x3x180x320xf32> loc(#loc171)
    %178 = "onnx.Add"(%177, %0) {onnx_node_name = "Sub_14-Initializer_398_30"} : (tensor<1x3x180x320xf32>, tensor<3x1x1xf32>) -> tensor<1x3x180x320xf32> loc(#loc476)
    %179 = "onnx.Div"(%178, %68) {onnx_node_name = "Div_16"} : (tensor<1x3x180x320xf32>, tensor<3x1x1xf32>) -> tensor<1x3x180x320xf32> loc(#loc174)
    %180 = "onnx.Conv"(%179, %76, %77) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_17",
      pads = [1, 1, 1, 1],
      strides = [2, 2]} : (tensor<1x3x180x320xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<1x16x90x160xf32> loc(#loc175)
    %181 = "onnx.Add"(%180, %71) {onnx_node_name = "Add_19"} : (tensor<1x16x90x160xf32>, tensor<f32>) -> tensor<1x16x90x160xf32> loc(#loc176)
    %182 = "onnx.Clip"(%181, %69, %70) {onnx_node_name = "Clip_22_12"} : (tensor<1x16x90x160xf32>, tensor<f32>, tensor<f32>) -> tensor<1x16x90x160xf32> loc(#loc177)
    %183 = "onnx.Div"(%182, %70) {onnx_node_name = "Div_24"} : (tensor<1x16x90x160xf32>, tensor<f32>) -> tensor<1x16x90x160xf32> loc(#loc178)
    %184 = "onnx.Mul"(%180, %183) {onnx_node_name = "Mul_25"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc179)
    %185 = "onnx.Conv"(%184, %78, %79) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 16 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_26",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x16x90x160xf32>, tensor<16x1x3x3xf32>, tensor<16xf32>) -> tensor<1x16x90x160xf32> loc(#loc180)
    %186 = "onnx.Relu"(%185) {onnx_node_name = "Relu_27"} : (tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc181)
    %187 = "onnx.Conv"(%186, %80, %81) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_28",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x16x90x160xf32>, tensor<16x16x1x1xf32>, tensor<16xf32>) -> tensor<1x16x90x160xf32> loc(#loc182)
    %188 = "onnx.Add"(%187, %184) {onnx_node_name = "Add_29"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc183)
    %189 = "onnx.Conv"(%188, %82, %83) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_30",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x16x90x160xf32>, tensor<64x16x1x1xf32>, tensor<64xf32>) -> tensor<1x64x90x160xf32> loc(#loc184)
    %190 = "onnx.Relu"(%189) {onnx_node_name = "Relu_31"} : (tensor<1x64x90x160xf32>) -> tensor<1x64x90x160xf32> loc(#loc185)
    %191 = "onnx.Conv"(%190, %84, %85) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 64 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_32",
      pads = [1, 1, 1, 1],
      strides = [2, 2]} : (tensor<1x64x90x160xf32>, tensor<64x1x3x3xf32>, tensor<64xf32>) -> tensor<1x64x45x80xf32> loc(#loc186)
    %192 = "onnx.Relu"(%191) {onnx_node_name = "Relu_33"} : (tensor<1x64x45x80xf32>) -> tensor<1x64x45x80xf32> loc(#loc187)
    %193 = "onnx.Conv"(%192, %86, %87) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_34",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x64x45x80xf32>, tensor<24x64x1x1xf32>, tensor<24xf32>) -> tensor<1x24x45x80xf32> loc(#loc188)
    %194 = "onnx.Conv"(%193, %88, %89) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_35",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x24x45x80xf32>, tensor<72x24x1x1xf32>, tensor<72xf32>) -> tensor<1x72x45x80xf32> loc(#loc189)
    %195 = "onnx.Relu"(%194) {onnx_node_name = "Relu_36"} : (tensor<1x72x45x80xf32>) -> tensor<1x72x45x80xf32> loc(#loc190)
    %196 = "onnx.Conv"(%195, %90, %91) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 72 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_37",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x72x45x80xf32>, tensor<72x1x3x3xf32>, tensor<72xf32>) -> tensor<1x72x45x80xf32> loc(#loc191)
    %197 = "onnx.Relu"(%196) {onnx_node_name = "Relu_38"} : (tensor<1x72x45x80xf32>) -> tensor<1x72x45x80xf32> loc(#loc192)
    %198 = "onnx.Conv"(%197, %92, %93) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_39",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x72x45x80xf32>, tensor<24x72x1x1xf32>, tensor<24xf32>) -> tensor<1x24x45x80xf32> loc(#loc193)
    %199 = "onnx.Add"(%198, %193) {onnx_node_name = "Add_40"} : (tensor<1x24x45x80xf32>, tensor<1x24x45x80xf32>) -> tensor<1x24x45x80xf32> loc(#loc194)
    %200 = "onnx.Conv"(%199, %94, %95) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_41",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x24x45x80xf32>, tensor<72x24x1x1xf32>, tensor<72xf32>) -> tensor<1x72x45x80xf32> loc(#loc195)
    %201 = "onnx.Relu"(%200) {onnx_node_name = "Relu_42"} : (tensor<1x72x45x80xf32>) -> tensor<1x72x45x80xf32> loc(#loc196)
    %202 = "onnx.Conv"(%201, %96, %97) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 72 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_43",
      pads = [2, 2, 2, 2],
      strides = [2, 2]} : (tensor<1x72x45x80xf32>, tensor<72x1x5x5xf32>, tensor<72xf32>) -> tensor<1x72x23x40xf32> loc(#loc197)
    %203 = "onnx.Relu"(%202) {onnx_node_name = "Relu_44"} : (tensor<1x72x23x40xf32>) -> tensor<1x72x23x40xf32> loc(#loc198)
    %204 = "onnx.ReduceMeanV13"(%203) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_45_45"} : (tensor<1x72x23x40xf32>) -> tensor<1x72x1x1xf32> loc(#loc199)
    %205 = "onnx.Conv"(%204, %146, %145) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_46",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x72x1x1xf32>, tensor<24x72x1x1xf32>, tensor<24xf32>) -> tensor<1x24x1x1xf32> loc(#loc200)
    %206 = "onnx.Relu"(%205) {onnx_node_name = "Relu_47"} : (tensor<1x24x1x1xf32>) -> tensor<1x24x1x1xf32> loc(#loc201)
    %207 = "onnx.Conv"(%206, %148, %147) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_48",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x24x1x1xf32>, tensor<72x24x1x1xf32>, tensor<72xf32>) -> tensor<1x72x1x1xf32> loc(#loc202)
    %208 = "onnx.Add"(%207, %71) {onnx_node_name = "Add_50"} : (tensor<1x72x1x1xf32>, tensor<f32>) -> tensor<1x72x1x1xf32> loc(#loc203)
    %209 = "onnx.Clip"(%208, %69, %70) {onnx_node_name = "Clip_53_32"} : (tensor<1x72x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x72x1x1xf32> loc(#loc204)
    %210 = "onnx.Div"(%209, %70) {onnx_node_name = "Div_55"} : (tensor<1x72x1x1xf32>, tensor<f32>) -> tensor<1x72x1x1xf32> loc(#loc205)
    %211 = "onnx.Mul"(%210, %203) {onnx_node_name = "Mul_56"} : (tensor<1x72x1x1xf32>, tensor<1x72x23x40xf32>) -> tensor<1x72x23x40xf32> loc(#loc206)
    %212 = "onnx.Conv"(%211, %98, %99) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_57",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x72x23x40xf32>, tensor<40x72x1x1xf32>, tensor<40xf32>) -> tensor<1x40x23x40xf32> loc(#loc207)
    %213 = "onnx.Conv"(%212, %100, %101) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_58",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x40x23x40xf32>, tensor<120x40x1x1xf32>, tensor<120xf32>) -> tensor<1x120x23x40xf32> loc(#loc208)
    %214 = "onnx.Relu"(%213) {onnx_node_name = "Relu_59"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc209)
    %215 = "onnx.Conv"(%214, %102, %103) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 120 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_60",
      pads = [2, 2, 2, 2],
      strides = [1, 1]} : (tensor<1x120x23x40xf32>, tensor<120x1x5x5xf32>, tensor<120xf32>) -> tensor<1x120x23x40xf32> loc(#loc210)
    %216 = "onnx.Relu"(%215) {onnx_node_name = "Relu_61"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc211)
    %217 = "onnx.ReduceMeanV13"(%216) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_62_51"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x1x1xf32> loc(#loc212)
    %218 = "onnx.Conv"(%217, %150, %149) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_63",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x120x1x1xf32>, tensor<32x120x1x1xf32>, tensor<32xf32>) -> tensor<1x32x1x1xf32> loc(#loc213)
    %219 = "onnx.Relu"(%218) {onnx_node_name = "Relu_64"} : (tensor<1x32x1x1xf32>) -> tensor<1x32x1x1xf32> loc(#loc214)
    %220 = "onnx.Conv"(%219, %152, %151) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_65",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x32x1x1xf32>, tensor<120x32x1x1xf32>, tensor<120xf32>) -> tensor<1x120x1x1xf32> loc(#loc215)
    %221 = "onnx.Add"(%220, %71) {onnx_node_name = "Add_67"} : (tensor<1x120x1x1xf32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc216)
    %222 = "onnx.Clip"(%221, %69, %70) {onnx_node_name = "Clip_70_4"} : (tensor<1x120x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc217)
    %223 = "onnx.Div"(%222, %70) {onnx_node_name = "Div_72"} : (tensor<1x120x1x1xf32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc218)
    %224 = "onnx.Mul"(%223, %216) {onnx_node_name = "Mul_73"} : (tensor<1x120x1x1xf32>, tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc219)
    %225 = "onnx.Conv"(%224, %104, %105) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_74",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x120x23x40xf32>, tensor<40x120x1x1xf32>, tensor<40xf32>) -> tensor<1x40x23x40xf32> loc(#loc220)
    %226 = "onnx.Add"(%225, %212) {onnx_node_name = "Add_75"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc221)
    %227 = "onnx.Conv"(%226, %106, %107) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_76",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x40x23x40xf32>, tensor<120x40x1x1xf32>, tensor<120xf32>) -> tensor<1x120x23x40xf32> loc(#loc222)
    %228 = "onnx.Relu"(%227) {onnx_node_name = "Relu_77"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc223)
    %229 = "onnx.Conv"(%228, %108, %109) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 120 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_78",
      pads = [2, 2, 2, 2],
      strides = [1, 1]} : (tensor<1x120x23x40xf32>, tensor<120x1x5x5xf32>, tensor<120xf32>) -> tensor<1x120x23x40xf32> loc(#loc224)
    %230 = "onnx.Relu"(%229) {onnx_node_name = "Relu_79"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc225)
    %231 = "onnx.ReduceMeanV13"(%230) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_80_18"} : (tensor<1x120x23x40xf32>) -> tensor<1x120x1x1xf32> loc(#loc226)
    %232 = "onnx.Conv"(%231, %154, %153) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_81",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x120x1x1xf32>, tensor<32x120x1x1xf32>, tensor<32xf32>) -> tensor<1x32x1x1xf32> loc(#loc227)
    %233 = "onnx.Relu"(%232) {onnx_node_name = "Relu_82"} : (tensor<1x32x1x1xf32>) -> tensor<1x32x1x1xf32> loc(#loc228)
    %234 = "onnx.Conv"(%233, %156, %155) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_83",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x32x1x1xf32>, tensor<120x32x1x1xf32>, tensor<120xf32>) -> tensor<1x120x1x1xf32> loc(#loc229)
    %235 = "onnx.Add"(%234, %71) {onnx_node_name = "Add_85"} : (tensor<1x120x1x1xf32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc230)
    %236 = "onnx.Clip"(%235, %69, %70) {onnx_node_name = "Clip_88_48"} : (tensor<1x120x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc231)
    %237 = "onnx.Div"(%236, %70) {onnx_node_name = "Div_90"} : (tensor<1x120x1x1xf32>, tensor<f32>) -> tensor<1x120x1x1xf32> loc(#loc232)
    %238 = "onnx.Mul"(%237, %230) {onnx_node_name = "Mul_91"} : (tensor<1x120x1x1xf32>, tensor<1x120x23x40xf32>) -> tensor<1x120x23x40xf32> loc(#loc233)
    %239 = "onnx.Conv"(%238, %110, %111) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_92",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x120x23x40xf32>, tensor<40x120x1x1xf32>, tensor<40xf32>) -> tensor<1x40x23x40xf32> loc(#loc234)
    %240 = "onnx.Add"(%239, %226) {onnx_node_name = "Add_93"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc235)
    %241 = "onnx.Conv"(%240, %112, %113) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_94",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x40x23x40xf32>, tensor<240x40x1x1xf32>, tensor<240xf32>) -> tensor<1x240x23x40xf32> loc(#loc236)
    %242 = "onnx.Add"(%241, %71) {onnx_node_name = "Add_96"} : (tensor<1x240x23x40xf32>, tensor<f32>) -> tensor<1x240x23x40xf32> loc(#loc237)
    %243 = "onnx.Clip"(%242, %69, %70) {onnx_node_name = "Clip_99_36"} : (tensor<1x240x23x40xf32>, tensor<f32>, tensor<f32>) -> tensor<1x240x23x40xf32> loc(#loc238)
    %244 = "onnx.Div"(%243, %70) {onnx_node_name = "Div_101"} : (tensor<1x240x23x40xf32>, tensor<f32>) -> tensor<1x240x23x40xf32> loc(#loc239)
    %245 = "onnx.Mul"(%241, %244) {onnx_node_name = "Mul_102"} : (tensor<1x240x23x40xf32>, tensor<1x240x23x40xf32>) -> tensor<1x240x23x40xf32> loc(#loc240)
    %246 = "onnx.Conv"(%245, %114, %115) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 240 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_103",
      pads = [1, 1, 1, 1],
      strides = [2, 2]} : (tensor<1x240x23x40xf32>, tensor<240x1x3x3xf32>, tensor<240xf32>) -> tensor<1x240x12x20xf32> loc(#loc241)
    %247 = "onnx.Add"(%246, %71) {onnx_node_name = "Add_105"} : (tensor<1x240x12x20xf32>, tensor<f32>) -> tensor<1x240x12x20xf32> loc(#loc242)
    %248 = "onnx.Clip"(%247, %69, %70) {onnx_node_name = "Clip_108_29"} : (tensor<1x240x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x240x12x20xf32> loc(#loc243)
    %249 = "onnx.Div"(%248, %70) {onnx_node_name = "Div_110"} : (tensor<1x240x12x20xf32>, tensor<f32>) -> tensor<1x240x12x20xf32> loc(#loc244)
    %250 = "onnx.Mul"(%246, %249) {onnx_node_name = "Mul_111"} : (tensor<1x240x12x20xf32>, tensor<1x240x12x20xf32>) -> tensor<1x240x12x20xf32> loc(#loc245)
    %251 = "onnx.Conv"(%250, %116, %117) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_112",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x240x12x20xf32>, tensor<80x240x1x1xf32>, tensor<80xf32>) -> tensor<1x80x12x20xf32> loc(#loc246)
    %252 = "onnx.Conv"(%251, %118, %119) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_113",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x80x12x20xf32>, tensor<200x80x1x1xf32>, tensor<200xf32>) -> tensor<1x200x12x20xf32> loc(#loc247)
    %253 = "onnx.Add"(%252, %71) {onnx_node_name = "Add_115"} : (tensor<1x200x12x20xf32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc248)
    %254 = "onnx.Clip"(%253, %69, %70) {onnx_node_name = "Clip_118_42"} : (tensor<1x200x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc249)
    %255 = "onnx.Div"(%254, %70) {onnx_node_name = "Div_120"} : (tensor<1x200x12x20xf32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc250)
    %256 = "onnx.Mul"(%252, %255) {onnx_node_name = "Mul_121"} : (tensor<1x200x12x20xf32>, tensor<1x200x12x20xf32>) -> tensor<1x200x12x20xf32> loc(#loc251)
    %257 = "onnx.Conv"(%256, %120, %121) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 200 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_122",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x200x12x20xf32>, tensor<200x1x3x3xf32>, tensor<200xf32>) -> tensor<1x200x12x20xf32> loc(#loc252)
    %258 = "onnx.Add"(%257, %71) {onnx_node_name = "Add_124"} : (tensor<1x200x12x20xf32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc253)
    %259 = "onnx.Clip"(%258, %69, %70) {onnx_node_name = "Clip_127_39"} : (tensor<1x200x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc254)
    %260 = "onnx.Div"(%259, %70) {onnx_node_name = "Div_129"} : (tensor<1x200x12x20xf32>, tensor<f32>) -> tensor<1x200x12x20xf32> loc(#loc255)
    %261 = "onnx.Mul"(%257, %260) {onnx_node_name = "Mul_130"} : (tensor<1x200x12x20xf32>, tensor<1x200x12x20xf32>) -> tensor<1x200x12x20xf32> loc(#loc256)
    %262 = "onnx.Conv"(%261, %122, %123) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_131",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x200x12x20xf32>, tensor<80x200x1x1xf32>, tensor<80xf32>) -> tensor<1x80x12x20xf32> loc(#loc257)
    %263 = "onnx.Add"(%262, %251) {onnx_node_name = "Add_132"} : (tensor<1x80x12x20xf32>, tensor<1x80x12x20xf32>) -> tensor<1x80x12x20xf32> loc(#loc258)
    %264 = "onnx.Conv"(%263, %7, %8) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_133",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x80x12x20xf32>, tensor<184x80x1x1xf32>, tensor<184xf32>) -> tensor<1x184x12x20xf32> loc(#loc259)
    %265 = "onnx.Add"(%264, %71) {onnx_node_name = "Add_135"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc260)
    %266 = "onnx.Clip"(%265, %69, %70) {onnx_node_name = "Clip_138_46"} : (tensor<1x184x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc261)
    %267 = "onnx.Div"(%266, %70) {onnx_node_name = "Div_140"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc262)
    %268 = "onnx.Mul"(%264, %267) {onnx_node_name = "Mul_141"} : (tensor<1x184x12x20xf32>, tensor<1x184x12x20xf32>) -> tensor<1x184x12x20xf32> loc(#loc263)
    %269 = "onnx.Conv"(%268, %9, %10) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 184 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_142",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x184x12x20xf32>, tensor<184x1x3x3xf32>, tensor<184xf32>) -> tensor<1x184x12x20xf32> loc(#loc264)
    %270 = "onnx.Add"(%269, %71) {onnx_node_name = "Add_144"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc265)
    %271 = "onnx.Clip"(%270, %69, %70) {onnx_node_name = "Clip_147_7"} : (tensor<1x184x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc266)
    %272 = "onnx.Div"(%271, %70) {onnx_node_name = "Div_149"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc267)
    %273 = "onnx.Mul"(%269, %272) {onnx_node_name = "Mul_150"} : (tensor<1x184x12x20xf32>, tensor<1x184x12x20xf32>) -> tensor<1x184x12x20xf32> loc(#loc268)
    %274 = "onnx.Conv"(%273, %11, %12) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_151",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x184x12x20xf32>, tensor<80x184x1x1xf32>, tensor<80xf32>) -> tensor<1x80x12x20xf32> loc(#loc269)
    %275 = "onnx.Add"(%274, %263) {onnx_node_name = "Add_152"} : (tensor<1x80x12x20xf32>, tensor<1x80x12x20xf32>) -> tensor<1x80x12x20xf32> loc(#loc270)
    %276 = "onnx.Conv"(%275, %13, %14) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_153",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x80x12x20xf32>, tensor<184x80x1x1xf32>, tensor<184xf32>) -> tensor<1x184x12x20xf32> loc(#loc271)
    %277 = "onnx.Add"(%276, %71) {onnx_node_name = "Add_155"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc272)
    %278 = "onnx.Clip"(%277, %69, %70) {onnx_node_name = "Clip_158_24"} : (tensor<1x184x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc273)
    %279 = "onnx.Div"(%278, %70) {onnx_node_name = "Div_160"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc274)
    %280 = "onnx.Mul"(%276, %279) {onnx_node_name = "Mul_161"} : (tensor<1x184x12x20xf32>, tensor<1x184x12x20xf32>) -> tensor<1x184x12x20xf32> loc(#loc275)
    %281 = "onnx.Conv"(%280, %15, %16) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 184 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_162",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x184x12x20xf32>, tensor<184x1x3x3xf32>, tensor<184xf32>) -> tensor<1x184x12x20xf32> loc(#loc276)
    %282 = "onnx.Add"(%281, %71) {onnx_node_name = "Add_164"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc277)
    %283 = "onnx.Clip"(%282, %69, %70) {onnx_node_name = "Clip_167_37"} : (tensor<1x184x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc278)
    %284 = "onnx.Div"(%283, %70) {onnx_node_name = "Div_169"} : (tensor<1x184x12x20xf32>, tensor<f32>) -> tensor<1x184x12x20xf32> loc(#loc279)
    %285 = "onnx.Mul"(%281, %284) {onnx_node_name = "Mul_170"} : (tensor<1x184x12x20xf32>, tensor<1x184x12x20xf32>) -> tensor<1x184x12x20xf32> loc(#loc280)
    %286 = "onnx.Conv"(%285, %17, %18) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_171",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x184x12x20xf32>, tensor<80x184x1x1xf32>, tensor<80xf32>) -> tensor<1x80x12x20xf32> loc(#loc281)
    %287 = "onnx.Add"(%286, %275) {onnx_node_name = "Add_172"} : (tensor<1x80x12x20xf32>, tensor<1x80x12x20xf32>) -> tensor<1x80x12x20xf32> loc(#loc282)
    %288 = "onnx.Conv"(%287, %19, %20) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_173",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x80x12x20xf32>, tensor<480x80x1x1xf32>, tensor<480xf32>) -> tensor<1x480x12x20xf32> loc(#loc283)
    %289 = "onnx.Add"(%288, %71) {onnx_node_name = "Add_175"} : (tensor<1x480x12x20xf32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc284)
    %290 = "onnx.Clip"(%289, %69, %70) {onnx_node_name = "Clip_178_50"} : (tensor<1x480x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc285)
    %291 = "onnx.Div"(%290, %70) {onnx_node_name = "Div_180"} : (tensor<1x480x12x20xf32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc286)
    %292 = "onnx.Mul"(%288, %291) {onnx_node_name = "Mul_181"} : (tensor<1x480x12x20xf32>, tensor<1x480x12x20xf32>) -> tensor<1x480x12x20xf32> loc(#loc287)
    %293 = "onnx.Conv"(%292, %21, %22) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 480 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_182",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x480x12x20xf32>, tensor<480x1x3x3xf32>, tensor<480xf32>) -> tensor<1x480x12x20xf32> loc(#loc288)
    %294 = "onnx.Add"(%293, %71) {onnx_node_name = "Add_184"} : (tensor<1x480x12x20xf32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc289)
    %295 = "onnx.Clip"(%294, %69, %70) {onnx_node_name = "Clip_187_0"} : (tensor<1x480x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc290)
    %296 = "onnx.Div"(%295, %70) {onnx_node_name = "Div_189"} : (tensor<1x480x12x20xf32>, tensor<f32>) -> tensor<1x480x12x20xf32> loc(#loc291)
    %297 = "onnx.Mul"(%293, %296) {onnx_node_name = "Mul_190"} : (tensor<1x480x12x20xf32>, tensor<1x480x12x20xf32>) -> tensor<1x480x12x20xf32> loc(#loc292)
    %298 = "onnx.ReduceMeanV13"(%297) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_191_26"} : (tensor<1x480x12x20xf32>) -> tensor<1x480x1x1xf32> loc(#loc293)
    %299 = "onnx.Conv"(%298, %126, %125) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_192",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x480x1x1xf32>, tensor<120x480x1x1xf32>, tensor<120xf32>) -> tensor<1x120x1x1xf32> loc(#loc294)
    %300 = "onnx.Relu"(%299) {onnx_node_name = "Relu_193"} : (tensor<1x120x1x1xf32>) -> tensor<1x120x1x1xf32> loc(#loc295)
    %301 = "onnx.Conv"(%300, %128, %127) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_194",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x120x1x1xf32>, tensor<480x120x1x1xf32>, tensor<480xf32>) -> tensor<1x480x1x1xf32> loc(#loc296)
    %302 = "onnx.Add"(%301, %71) {onnx_node_name = "Add_196"} : (tensor<1x480x1x1xf32>, tensor<f32>) -> tensor<1x480x1x1xf32> loc(#loc297)
    %303 = "onnx.Clip"(%302, %69, %70) {onnx_node_name = "Clip_199_43"} : (tensor<1x480x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x480x1x1xf32> loc(#loc298)
    %304 = "onnx.Div"(%303, %70) {onnx_node_name = "Div_201"} : (tensor<1x480x1x1xf32>, tensor<f32>) -> tensor<1x480x1x1xf32> loc(#loc299)
    %305 = "onnx.Mul"(%304, %297) {onnx_node_name = "Mul_202"} : (tensor<1x480x1x1xf32>, tensor<1x480x12x20xf32>) -> tensor<1x480x12x20xf32> loc(#loc300)
    %306 = "onnx.Conv"(%305, %23, %24) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_203",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x480x12x20xf32>, tensor<112x480x1x1xf32>, tensor<112xf32>) -> tensor<1x112x12x20xf32> loc(#loc301)
    %307 = "onnx.Conv"(%306, %25, %26) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_204",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x112x12x20xf32>, tensor<672x112x1x1xf32>, tensor<672xf32>) -> tensor<1x672x12x20xf32> loc(#loc302)
    %308 = "onnx.Add"(%307, %71) {onnx_node_name = "Add_206"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc303)
    %309 = "onnx.Clip"(%308, %69, %70) {onnx_node_name = "Clip_209_28"} : (tensor<1x672x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc304)
    %310 = "onnx.Div"(%309, %70) {onnx_node_name = "Div_211"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc305)
    %311 = "onnx.Mul"(%307, %310) {onnx_node_name = "Mul_212"} : (tensor<1x672x12x20xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc306)
    %312 = "onnx.Conv"(%311, %27, %28) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 672 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_213",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x672x12x20xf32>, tensor<672x1x3x3xf32>, tensor<672xf32>) -> tensor<1x672x12x20xf32> loc(#loc307)
    %313 = "onnx.Add"(%312, %71) {onnx_node_name = "Add_215"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc308)
    %314 = "onnx.Clip"(%313, %69, %70) {onnx_node_name = "Clip_218_25"} : (tensor<1x672x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc309)
    %315 = "onnx.Div"(%314, %70) {onnx_node_name = "Div_220"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc310)
    %316 = "onnx.Mul"(%312, %315) {onnx_node_name = "Mul_221"} : (tensor<1x672x12x20xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc311)
    %317 = "onnx.ReduceMeanV13"(%316) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_222_47"} : (tensor<1x672x12x20xf32>) -> tensor<1x672x1x1xf32> loc(#loc312)
    %318 = "onnx.Conv"(%317, %130, %129) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_223",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x672x1x1xf32>, tensor<168x672x1x1xf32>, tensor<168xf32>) -> tensor<1x168x1x1xf32> loc(#loc313)
    %319 = "onnx.Relu"(%318) {onnx_node_name = "Relu_224"} : (tensor<1x168x1x1xf32>) -> tensor<1x168x1x1xf32> loc(#loc314)
    %320 = "onnx.Conv"(%319, %132, %131) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_225",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x168x1x1xf32>, tensor<672x168x1x1xf32>, tensor<672xf32>) -> tensor<1x672x1x1xf32> loc(#loc315)
    %321 = "onnx.Add"(%320, %71) {onnx_node_name = "Add_227"} : (tensor<1x672x1x1xf32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc316)
    %322 = "onnx.Clip"(%321, %69, %70) {onnx_node_name = "Clip_230_27"} : (tensor<1x672x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc317)
    %323 = "onnx.Div"(%322, %70) {onnx_node_name = "Div_232"} : (tensor<1x672x1x1xf32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc318)
    %324 = "onnx.Mul"(%323, %316) {onnx_node_name = "Mul_233"} : (tensor<1x672x1x1xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc319)
    %325 = "onnx.Conv"(%324, %29, %30) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_234",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x672x12x20xf32>, tensor<112x672x1x1xf32>, tensor<112xf32>) -> tensor<1x112x12x20xf32> loc(#loc320)
    %326 = "onnx.Add"(%325, %306) {onnx_node_name = "Add_235"} : (tensor<1x112x12x20xf32>, tensor<1x112x12x20xf32>) -> tensor<1x112x12x20xf32> loc(#loc321)
    %327 = "onnx.Conv"(%326, %31, %32) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_236",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x112x12x20xf32>, tensor<672x112x1x1xf32>, tensor<672xf32>) -> tensor<1x672x12x20xf32> loc(#loc322)
    %328 = "onnx.Add"(%327, %71) {onnx_node_name = "Add_238"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc323)
    %329 = "onnx.Clip"(%328, %69, %70) {onnx_node_name = "Clip_241_35"} : (tensor<1x672x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc324)
    %330 = "onnx.Div"(%329, %70) {onnx_node_name = "Div_243"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc325)
    %331 = "onnx.Mul"(%327, %330) {onnx_node_name = "Mul_244"} : (tensor<1x672x12x20xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc326)
    %332 = "onnx.Conv"(%331, %33, %34) {
      auto_pad = "NOTSET",
      dilations = [2, 2],
      group = 672 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_245",
      pads = [4, 4, 4, 4],
      strides = [1, 1]} : (tensor<1x672x12x20xf32>, tensor<672x1x5x5xf32>, tensor<672xf32>) -> tensor<1x672x12x20xf32> loc(#loc327)
    %333 = "onnx.Add"(%332, %71) {onnx_node_name = "Add_247"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc328)
    %334 = "onnx.Clip"(%333, %69, %70) {onnx_node_name = "Clip_250_40"} : (tensor<1x672x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc329)
    %335 = "onnx.Div"(%334, %70) {onnx_node_name = "Div_252"} : (tensor<1x672x12x20xf32>, tensor<f32>) -> tensor<1x672x12x20xf32> loc(#loc330)
    %336 = "onnx.Mul"(%332, %335) {onnx_node_name = "Mul_253"} : (tensor<1x672x12x20xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc331)
    %337 = "onnx.ReduceMeanV13"(%336) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_254_19"} : (tensor<1x672x12x20xf32>) -> tensor<1x672x1x1xf32> loc(#loc332)
    %338 = "onnx.Conv"(%337, %134, %133) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_255",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x672x1x1xf32>, tensor<168x672x1x1xf32>, tensor<168xf32>) -> tensor<1x168x1x1xf32> loc(#loc333)
    %339 = "onnx.Relu"(%338) {onnx_node_name = "Relu_256"} : (tensor<1x168x1x1xf32>) -> tensor<1x168x1x1xf32> loc(#loc334)
    %340 = "onnx.Conv"(%339, %136, %135) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_257",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x168x1x1xf32>, tensor<672x168x1x1xf32>, tensor<672xf32>) -> tensor<1x672x1x1xf32> loc(#loc335)
    %341 = "onnx.Add"(%340, %71) {onnx_node_name = "Add_259"} : (tensor<1x672x1x1xf32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc336)
    %342 = "onnx.Clip"(%341, %69, %70) {onnx_node_name = "Clip_262_49"} : (tensor<1x672x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc337)
    %343 = "onnx.Div"(%342, %70) {onnx_node_name = "Div_264"} : (tensor<1x672x1x1xf32>, tensor<f32>) -> tensor<1x672x1x1xf32> loc(#loc338)
    %344 = "onnx.Mul"(%343, %336) {onnx_node_name = "Mul_265"} : (tensor<1x672x1x1xf32>, tensor<1x672x12x20xf32>) -> tensor<1x672x12x20xf32> loc(#loc339)
    %345 = "onnx.Conv"(%344, %35, %36) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_266",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x672x12x20xf32>, tensor<160x672x1x1xf32>, tensor<160xf32>) -> tensor<1x160x12x20xf32> loc(#loc340)
    %346 = "onnx.Conv"(%345, %37, %38) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_267",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x160x12x20xf32>, tensor<960x160x1x1xf32>, tensor<960xf32>) -> tensor<1x960x12x20xf32> loc(#loc341)
    %347 = "onnx.Add"(%346, %71) {onnx_node_name = "Add_269"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc342)
    %348 = "onnx.Clip"(%347, %69, %70) {onnx_node_name = "Clip_272_20"} : (tensor<1x960x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc343)
    %349 = "onnx.Div"(%348, %70) {onnx_node_name = "Div_274"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc344)
    %350 = "onnx.Mul"(%346, %349) {onnx_node_name = "Mul_275"} : (tensor<1x960x12x20xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc345)
    %351 = "onnx.Conv"(%350, %39, %40) {
      auto_pad = "NOTSET",
      dilations = [2, 2],
      group = 960 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_276",
      pads = [4, 4, 4, 4],
      strides = [1, 1]} : (tensor<1x960x12x20xf32>, tensor<960x1x5x5xf32>, tensor<960xf32>) -> tensor<1x960x12x20xf32> loc(#loc346)
    %352 = "onnx.Add"(%351, %71) {onnx_node_name = "Add_278"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc347)
    %353 = "onnx.Clip"(%352, %69, %70) {onnx_node_name = "Clip_281_2"} : (tensor<1x960x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc348)
    %354 = "onnx.Div"(%353, %70) {onnx_node_name = "Div_283"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc349)
    %355 = "onnx.Mul"(%351, %354) {onnx_node_name = "Mul_284"} : (tensor<1x960x12x20xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc350)
    %356 = "onnx.ReduceMeanV13"(%355) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_285_9"} : (tensor<1x960x12x20xf32>) -> tensor<1x960x1x1xf32> loc(#loc351)
    %357 = "onnx.Conv"(%356, %138, %137) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_286",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x1x1xf32>, tensor<240x960x1x1xf32>, tensor<240xf32>) -> tensor<1x240x1x1xf32> loc(#loc352)
    %358 = "onnx.Relu"(%357) {onnx_node_name = "Relu_287"} : (tensor<1x240x1x1xf32>) -> tensor<1x240x1x1xf32> loc(#loc353)
    %359 = "onnx.Conv"(%358, %140, %139) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_288",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x240x1x1xf32>, tensor<960x240x1x1xf32>, tensor<960xf32>) -> tensor<1x960x1x1xf32> loc(#loc354)
    %360 = "onnx.Add"(%359, %71) {onnx_node_name = "Add_290"} : (tensor<1x960x1x1xf32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc355)
    %361 = "onnx.Clip"(%360, %69, %70) {onnx_node_name = "Clip_293_44"} : (tensor<1x960x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc356)
    %362 = "onnx.Div"(%361, %70) {onnx_node_name = "Div_295"} : (tensor<1x960x1x1xf32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc357)
    %363 = "onnx.Mul"(%362, %355) {onnx_node_name = "Mul_296"} : (tensor<1x960x1x1xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc358)
    %364 = "onnx.Conv"(%363, %41, %42) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_297",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x12x20xf32>, tensor<160x960x1x1xf32>, tensor<160xf32>) -> tensor<1x160x12x20xf32> loc(#loc359)
    %365 = "onnx.Add"(%364, %345) {onnx_node_name = "Add_298"} : (tensor<1x160x12x20xf32>, tensor<1x160x12x20xf32>) -> tensor<1x160x12x20xf32> loc(#loc360)
    %366 = "onnx.Conv"(%365, %43, %44) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_299",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x160x12x20xf32>, tensor<960x160x1x1xf32>, tensor<960xf32>) -> tensor<1x960x12x20xf32> loc(#loc361)
    %367 = "onnx.Add"(%366, %71) {onnx_node_name = "Add_301"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc362)
    %368 = "onnx.Clip"(%367, %69, %70) {onnx_node_name = "Clip_304_21"} : (tensor<1x960x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc363)
    %369 = "onnx.Div"(%368, %70) {onnx_node_name = "Div_306"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc364)
    %370 = "onnx.Mul"(%366, %369) {onnx_node_name = "Mul_307"} : (tensor<1x960x12x20xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc365)
    %371 = "onnx.Conv"(%370, %45, %46) {
      auto_pad = "NOTSET",
      dilations = [2, 2],
      group = 960 : si64,
      kernel_shape = [5, 5],
      onnx_node_name = "Conv_308",
      pads = [4, 4, 4, 4],
      strides = [1, 1]} : (tensor<1x960x12x20xf32>, tensor<960x1x5x5xf32>, tensor<960xf32>) -> tensor<1x960x12x20xf32> loc(#loc366)
    %372 = "onnx.Add"(%371, %71) {onnx_node_name = "Add_310"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc367)
    %373 = "onnx.Clip"(%372, %69, %70) {onnx_node_name = "Clip_313_22"} : (tensor<1x960x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc368)
    %374 = "onnx.Div"(%373, %70) {onnx_node_name = "Div_315"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc369)
    %375 = "onnx.Mul"(%371, %374) {onnx_node_name = "Mul_316"} : (tensor<1x960x12x20xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc370)
    %376 = "onnx.ReduceMeanV13"(%375) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_317_8"} : (tensor<1x960x12x20xf32>) -> tensor<1x960x1x1xf32> loc(#loc371)
    %377 = "onnx.Conv"(%376, %142, %141) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_318",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x1x1xf32>, tensor<240x960x1x1xf32>, tensor<240xf32>) -> tensor<1x240x1x1xf32> loc(#loc372)
    %378 = "onnx.Relu"(%377) {onnx_node_name = "Relu_319"} : (tensor<1x240x1x1xf32>) -> tensor<1x240x1x1xf32> loc(#loc373)
    %379 = "onnx.Conv"(%378, %144, %143) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_320",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x240x1x1xf32>, tensor<960x240x1x1xf32>, tensor<960xf32>) -> tensor<1x960x1x1xf32> loc(#loc374)
    %380 = "onnx.Add"(%379, %71) {onnx_node_name = "Add_322"} : (tensor<1x960x1x1xf32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc375)
    %381 = "onnx.Clip"(%380, %69, %70) {onnx_node_name = "Clip_325_3"} : (tensor<1x960x1x1xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc376)
    %382 = "onnx.Div"(%381, %70) {onnx_node_name = "Div_327"} : (tensor<1x960x1x1xf32>, tensor<f32>) -> tensor<1x960x1x1xf32> loc(#loc377)
    %383 = "onnx.Mul"(%382, %375) {onnx_node_name = "Mul_328"} : (tensor<1x960x1x1xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc378)
    %384 = "onnx.Conv"(%383, %47, %48) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_329",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x12x20xf32>, tensor<160x960x1x1xf32>, tensor<160xf32>) -> tensor<1x160x12x20xf32> loc(#loc379)
    %385 = "onnx.Add"(%384, %365) {onnx_node_name = "Add_330"} : (tensor<1x160x12x20xf32>, tensor<1x160x12x20xf32>) -> tensor<1x160x12x20xf32> loc(#loc380)
    %386 = "onnx.Conv"(%385, %49, %50) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_331",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x160x12x20xf32>, tensor<960x160x1x1xf32>, tensor<960xf32>) -> tensor<1x960x12x20xf32> loc(#loc381)
    %387 = "onnx.Add"(%386, %71) {onnx_node_name = "Add_333"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc382)
    %388 = "onnx.Clip"(%387, %69, %70) {onnx_node_name = "Clip_336_53"} : (tensor<1x960x12x20xf32>, tensor<f32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc383)
    %389 = "onnx.Div"(%388, %70) {onnx_node_name = "Div_338"} : (tensor<1x960x12x20xf32>, tensor<f32>) -> tensor<1x960x12x20xf32> loc(#loc384)
    %390 = "onnx.Mul"(%386, %389) {onnx_node_name = "Mul_339"} : (tensor<1x960x12x20xf32>, tensor<1x960x12x20xf32>) -> tensor<1x960x12x20xf32> loc(#loc385)
    %391 = "onnx.ReduceMeanV13"(%390) {
      axes = [2, 3],
      keepdims = 1 : si64,
      onnx_node_name = "GlobalAveragePool_342_14"} : (tensor<1x960x12x20xf32>) -> tensor<1x960x1x1xf32> loc(#loc386)
    %392 = "onnx.Conv"(%391, %124, %6) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_343",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x1x1xf32>, tensor<128x960x1x1xf32>, none) -> tensor<1x128x1x1xf32> loc(#loc387)
    %393 = "onnx.Sigmoid"(%392) {onnx_node_name = "Sigmoid_344"} : (tensor<1x128x1x1xf32>) -> tensor<1x128x1x1xf32> loc(#loc388)
    %394 = "onnx.Conv"(%390, %51, %52) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_340",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x960x12x20xf32>, tensor<128x960x1x1xf32>, tensor<128xf32>) -> tensor<1x128x12x20xf32> loc(#loc389)
    %395 = "onnx.Relu"(%394) {onnx_node_name = "Relu_341"} : (tensor<1x128x12x20xf32>) -> tensor<1x128x12x20xf32> loc(#loc390)
    %396 = "onnx.Mul"(%395, %393) {onnx_node_name = "Mul_345"} : (tensor<1x128x12x20xf32>, tensor<1x128x1x1xf32>) -> tensor<1x128x12x20xf32> loc(#loc391)
    %397:2 = "onnx.Split"(%396, %5) {axis = 1 : si64, onnx_node_name = "Split_349_17"} : (tensor<1x128x12x20xf32>, tensor<2xi64>) -> (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) loc(#loc392)
    %398 = "onnx.Concat"(%397#1, %arg4) {axis = 1 : si64, onnx_node_name = "Concat_350"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x128x12x20xf32> loc(#loc393)
    %399 = "onnx.Conv"(%398, %172, %171) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_351",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x128x12x20xf32>, tensor<128x128x3x3xf32>, tensor<128xf32>) -> tensor<1x128x12x20xf32> loc(#loc394)
    %400 = "onnx.Sigmoid"(%399) {onnx_node_name = "Sigmoid_352"} : (tensor<1x128x12x20xf32>) -> tensor<1x128x12x20xf32> loc(#loc395)
    %401:2 = "onnx.Split"(%400, %5) {axis = 1 : si64, onnx_node_name = "Split_353_38"} : (tensor<1x128x12x20xf32>, tensor<2xi64>) -> (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) loc(#loc396)
    %402 = "onnx.Mul"(%401#0, %arg4) {onnx_node_name = "Mul_354"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc397)
    %403 = "onnx.Concat"(%397#1, %402) {axis = 1 : si64, onnx_node_name = "Concat_355"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x128x12x20xf32> loc(#loc398)
    %404 = "onnx.Conv"(%403, %170, %169) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_356",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x128x12x20xf32>, tensor<64x128x3x3xf32>, tensor<64xf32>) -> tensor<1x64x12x20xf32> loc(#loc399)
    %405 = "onnx.Tanh"(%404) {onnx_node_name = "Tanh_357"} : (tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc400)
    %406 = "onnx.Mul"(%401#1, %405) {onnx_node_name = "Mul_361"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc401)
    %407 = "onnx.Sub"(%75, %401#1) {onnx_node_name = "Sub_359"} : (tensor<f32>, tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc402)
    %408 = "onnx.Mul"(%407, %arg4) {onnx_node_name = "Mul_360"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc403)
    %409 = "onnx.Add"(%408, %406) {onnx_node_name = "Add_362"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x64x12x20xf32> loc(#loc404)
    %410 = "onnx.Concat"(%397#0, %409) {axis = 1 : si64, onnx_node_name = "Concat_363"} : (tensor<1x64x12x20xf32>, tensor<1x64x12x20xf32>) -> tensor<1x128x12x20xf32> loc(#loc405)
    %411 = "onnx.Resize"(%410, %6, %64, %6) {
      antialias = 0 : si64,
      coordinate_transformation_mode = "pytorch_half_pixel",
      cubic_coeff_a = -7.500000e-01 : f32,
      exclude_outside = 0 : si64,
      extrapolation_value = 0.000000e+00 : f32,
      keep_aspect_ratio_policy = "stretch",
      mode = "linear",
      nearest_mode = "floor",
      onnx_node_name = "Resize_365_5"} : (tensor<1x128x12x20xf32>, none, tensor<4xf32>, none) -> tensor<1x128x24x40xf32> loc(#loc406)
    %412 = "onnx.Slice"(%411, %67, %72, %63, %74) {onnx_node_name = "Slice_371"} : (tensor<1x128x24x40xf32>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>) -> tensor<1x128x23x40xf32> loc(#loc407)
    %413 = "onnx.AveragePool"(%177) {
      auto_pad = "NOTSET",
      ceil_mode = 1 : si64,
      count_include_pad = 0 : si64,
      kernel_shape = [2, 2],
      onnx_node_name = "AveragePool_346",
      pads = [0, 0, 0, 0],
      strides = [2, 2]} : (tensor<1x3x180x320xf32>) -> tensor<1x3x90x160xf32> loc(#loc408)
    %414 = "onnx.AveragePool"(%413) {
      auto_pad = "NOTSET",
      ceil_mode = 1 : si64,
      count_include_pad = 0 : si64,
      kernel_shape = [2, 2],
      onnx_node_name = "AveragePool_347",
      pads = [0, 0, 0, 0],
      strides = [2, 2]} : (tensor<1x3x90x160xf32>) -> tensor<1x3x45x80xf32> loc(#loc409)
    %415 = "onnx.AveragePool"(%414) {
      auto_pad = "NOTSET",
      ceil_mode = 1 : si64,
      count_include_pad = 0 : si64,
      kernel_shape = [2, 2],
      onnx_node_name = "AveragePool_348",
      pads = [0, 0, 0, 0],
      strides = [2, 2]} : (tensor<1x3x45x80xf32>) -> tensor<1x3x23x40xf32> loc(#loc410)
    %416 = "onnx.Concat"(%412, %240, %415) {axis = 1 : si64, onnx_node_name = "Concat_372"} : (tensor<1x128x23x40xf32>, tensor<1x40x23x40xf32>, tensor<1x3x23x40xf32>) -> tensor<1x171x23x40xf32> loc(#loc411)
    %417 = "onnx.Conv"(%416, %53, %54) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_373",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x171x23x40xf32>, tensor<80x171x3x3xf32>, tensor<80xf32>) -> tensor<1x80x23x40xf32> loc(#loc412)
    %418 = "onnx.Relu"(%417) {onnx_node_name = "Relu_374"} : (tensor<1x80x23x40xf32>) -> tensor<1x80x23x40xf32> loc(#loc413)
    %419:2 = "onnx.Split"(%418, %4) {axis = 1 : si64, onnx_node_name = "Split_375_52"} : (tensor<1x80x23x40xf32>, tensor<2xi64>) -> (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) loc(#loc414)
    %420 = "onnx.Concat"(%419#1, %arg3) {axis = 1 : si64, onnx_node_name = "Concat_376"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x80x23x40xf32> loc(#loc415)
    %421 = "onnx.Conv"(%420, %168, %167) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_377",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x80x23x40xf32>, tensor<80x80x3x3xf32>, tensor<80xf32>) -> tensor<1x80x23x40xf32> loc(#loc416)
    %422 = "onnx.Sigmoid"(%421) {onnx_node_name = "Sigmoid_378"} : (tensor<1x80x23x40xf32>) -> tensor<1x80x23x40xf32> loc(#loc417)
    %423:2 = "onnx.Split"(%422, %4) {axis = 1 : si64, onnx_node_name = "Split_379_31"} : (tensor<1x80x23x40xf32>, tensor<2xi64>) -> (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) loc(#loc418)
    %424 = "onnx.Mul"(%423#0, %arg3) {onnx_node_name = "Mul_380"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc419)
    %425 = "onnx.Concat"(%419#1, %424) {axis = 1 : si64, onnx_node_name = "Concat_381"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x80x23x40xf32> loc(#loc420)
    %426 = "onnx.Conv"(%425, %166, %165) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_382",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x80x23x40xf32>, tensor<40x80x3x3xf32>, tensor<40xf32>) -> tensor<1x40x23x40xf32> loc(#loc421)
    %427 = "onnx.Tanh"(%426) {onnx_node_name = "Tanh_383"} : (tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc422)
    %428 = "onnx.Mul"(%423#1, %427) {onnx_node_name = "Mul_387"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc423)
    %429 = "onnx.Sub"(%75, %423#1) {onnx_node_name = "Sub_385"} : (tensor<f32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc424)
    %430 = "onnx.Mul"(%429, %arg3) {onnx_node_name = "Mul_386"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc425)
    %431 = "onnx.Add"(%430, %428) {onnx_node_name = "Add_388"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x40x23x40xf32> loc(#loc426)
    %432 = "onnx.Concat"(%419#0, %431) {axis = 1 : si64, onnx_node_name = "Concat_389"} : (tensor<1x40x23x40xf32>, tensor<1x40x23x40xf32>) -> tensor<1x80x23x40xf32> loc(#loc427)
    %433 = "onnx.Resize"(%432, %6, %64, %6) {
      antialias = 0 : si64,
      coordinate_transformation_mode = "pytorch_half_pixel",
      cubic_coeff_a = -7.500000e-01 : f32,
      exclude_outside = 0 : si64,
      extrapolation_value = 0.000000e+00 : f32,
      keep_aspect_ratio_policy = "stretch",
      mode = "linear",
      nearest_mode = "floor",
      onnx_node_name = "Resize_391_10"} : (tensor<1x80x23x40xf32>, none, tensor<4xf32>, none) -> tensor<1x80x46x80xf32> loc(#loc428)
    %434 = "onnx.Slice"(%433, %67, %73, %63, %74) {onnx_node_name = "Slice_397"} : (tensor<1x80x46x80xf32>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>, tensor<1xi64>) -> tensor<1x80x45x80xf32> loc(#loc429)
    %435 = "onnx.Concat"(%434, %199, %414) {axis = 1 : si64, onnx_node_name = "Concat_398"} : (tensor<1x80x45x80xf32>, tensor<1x24x45x80xf32>, tensor<1x3x45x80xf32>) -> tensor<1x107x45x80xf32> loc(#loc430)
    %436 = "onnx.Conv"(%435, %55, %56) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_399",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x107x45x80xf32>, tensor<40x107x3x3xf32>, tensor<40xf32>) -> tensor<1x40x45x80xf32> loc(#loc431)
    %437 = "onnx.Relu"(%436) {onnx_node_name = "Relu_400"} : (tensor<1x40x45x80xf32>) -> tensor<1x40x45x80xf32> loc(#loc432)
    %438:2 = "onnx.Split"(%437, %3) {axis = 1 : si64, onnx_node_name = "Split_401_41"} : (tensor<1x40x45x80xf32>, tensor<2xi64>) -> (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) loc(#loc433)
    %439 = "onnx.Concat"(%438#1, %arg2) {axis = 1 : si64, onnx_node_name = "Concat_402"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x40x45x80xf32> loc(#loc434)
    %440 = "onnx.Conv"(%439, %164, %163) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_403",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x40x45x80xf32>, tensor<40x40x3x3xf32>, tensor<40xf32>) -> tensor<1x40x45x80xf32> loc(#loc435)
    %441 = "onnx.Sigmoid"(%440) {onnx_node_name = "Sigmoid_404"} : (tensor<1x40x45x80xf32>) -> tensor<1x40x45x80xf32> loc(#loc436)
    %442:2 = "onnx.Split"(%441, %3) {axis = 1 : si64, onnx_node_name = "Split_405_6"} : (tensor<1x40x45x80xf32>, tensor<2xi64>) -> (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) loc(#loc437)
    %443 = "onnx.Mul"(%442#0, %arg2) {onnx_node_name = "Mul_406"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc438)
    %444 = "onnx.Concat"(%438#1, %443) {axis = 1 : si64, onnx_node_name = "Concat_407"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x40x45x80xf32> loc(#loc439)
    %445 = "onnx.Conv"(%444, %162, %161) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_408",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x40x45x80xf32>, tensor<20x40x3x3xf32>, tensor<20xf32>) -> tensor<1x20x45x80xf32> loc(#loc440)
    %446 = "onnx.Tanh"(%445) {onnx_node_name = "Tanh_409"} : (tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc441)
    %447 = "onnx.Mul"(%442#1, %446) {onnx_node_name = "Mul_413"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc442)
    %448 = "onnx.Sub"(%75, %442#1) {onnx_node_name = "Sub_411"} : (tensor<f32>, tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc443)
    %449 = "onnx.Mul"(%448, %arg2) {onnx_node_name = "Mul_412"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc444)
    %450 = "onnx.Add"(%449, %447) {onnx_node_name = "Add_414"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x20x45x80xf32> loc(#loc445)
    %451 = "onnx.Concat"(%438#0, %450) {axis = 1 : si64, onnx_node_name = "Concat_415"} : (tensor<1x20x45x80xf32>, tensor<1x20x45x80xf32>) -> tensor<1x40x45x80xf32> loc(#loc446)
    %452 = "onnx.Resize"(%451, %6, %64, %6) {
      antialias = 0 : si64,
      coordinate_transformation_mode = "pytorch_half_pixel",
      cubic_coeff_a = -7.500000e-01 : f32,
      exclude_outside = 0 : si64,
      extrapolation_value = 0.000000e+00 : f32,
      keep_aspect_ratio_policy = "stretch",
      mode = "linear",
      nearest_mode = "floor",
      onnx_node_name = "Resize_417_15"} : (tensor<1x40x45x80xf32>, none, tensor<4xf32>, none) -> tensor<1x40x90x160xf32> loc(#loc447)
    %453 = "onnx.Concat"(%452, %188, %413) {axis = 1 : si64, onnx_node_name = "Concat_418"} : (tensor<1x40x90x160xf32>, tensor<1x16x90x160xf32>, tensor<1x3x90x160xf32>) -> tensor<1x59x90x160xf32> loc(#loc448)
    %454 = "onnx.Conv"(%453, %57, %58) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_419",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x59x90x160xf32>, tensor<32x59x3x3xf32>, tensor<32xf32>) -> tensor<1x32x90x160xf32> loc(#loc449)
    %455 = "onnx.Relu"(%454) {onnx_node_name = "Relu_420"} : (tensor<1x32x90x160xf32>) -> tensor<1x32x90x160xf32> loc(#loc450)
    %456:2 = "onnx.Split"(%455, %2) {axis = 1 : si64, onnx_node_name = "Split_421_16"} : (tensor<1x32x90x160xf32>, tensor<2xi64>) -> (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) loc(#loc451)
    %457 = "onnx.Concat"(%456#1, %arg1) {axis = 1 : si64, onnx_node_name = "Concat_422"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x32x90x160xf32> loc(#loc452)
    %458 = "onnx.Conv"(%457, %160, %159) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_423",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x32x90x160xf32>, tensor<32x32x3x3xf32>, tensor<32xf32>) -> tensor<1x32x90x160xf32> loc(#loc453)
    %459 = "onnx.Sigmoid"(%458) {onnx_node_name = "Sigmoid_424"} : (tensor<1x32x90x160xf32>) -> tensor<1x32x90x160xf32> loc(#loc454)
    %460:2 = "onnx.Split"(%459, %2) {axis = 1 : si64, onnx_node_name = "Split_425_23"} : (tensor<1x32x90x160xf32>, tensor<2xi64>) -> (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) loc(#loc455)
    %461 = "onnx.Mul"(%460#0, %arg1) {onnx_node_name = "Mul_426"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc456)
    %462 = "onnx.Concat"(%456#1, %461) {axis = 1 : si64, onnx_node_name = "Concat_427"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x32x90x160xf32> loc(#loc457)
    %463 = "onnx.Conv"(%462, %158, %157) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_428",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x32x90x160xf32>, tensor<16x32x3x3xf32>, tensor<16xf32>) -> tensor<1x16x90x160xf32> loc(#loc458)
    %464 = "onnx.Tanh"(%463) {onnx_node_name = "Tanh_429"} : (tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc459)
    %465 = "onnx.Mul"(%460#1, %464) {onnx_node_name = "Mul_433"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc460)
    %466 = "onnx.Sub"(%75, %460#1) {onnx_node_name = "Sub_431"} : (tensor<f32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc461)
    %467 = "onnx.Mul"(%466, %arg1) {onnx_node_name = "Mul_432"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc462)
    %468 = "onnx.Add"(%467, %465) {onnx_node_name = "Add_434"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x16x90x160xf32> loc(#loc463)
    %469 = "onnx.Concat"(%456#0, %468) {axis = 1 : si64, onnx_node_name = "Concat_435"} : (tensor<1x16x90x160xf32>, tensor<1x16x90x160xf32>) -> tensor<1x32x90x160xf32> loc(#loc464)
    %470 = "onnx.Resize"(%469, %6, %64, %6) {
      antialias = 0 : si64,
      coordinate_transformation_mode = "pytorch_half_pixel",
      cubic_coeff_a = -7.500000e-01 : f32,
      exclude_outside = 0 : si64,
      extrapolation_value = 0.000000e+00 : f32,
      keep_aspect_ratio_policy = "stretch",
      mode = "linear",
      nearest_mode = "floor",
      onnx_node_name = "Resize_437_11"} : (tensor<1x32x90x160xf32>, none, tensor<4xf32>, none) -> tensor<1x32x180x320xf32> loc(#loc465)
    %471 = "onnx.Concat"(%470, %177) {axis = 1 : si64, onnx_node_name = "Concat_438"} : (tensor<1x32x180x320xf32>, tensor<1x3x180x320xf32>) -> tensor<1x35x180x320xf32> loc(#loc466)
    %472 = "onnx.Conv"(%471, %59, %60) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_439",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x35x180x320xf32>, tensor<16x35x3x3xf32>, tensor<16xf32>) -> tensor<1x16x180x320xf32> loc(#loc467)
    %473 = "onnx.Relu"(%472) {onnx_node_name = "Relu_440"} : (tensor<1x16x180x320xf32>) -> tensor<1x16x180x320xf32> loc(#loc468)
    %474 = "onnx.Conv"(%473, %61, %62) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [3, 3],
      onnx_node_name = "Conv_441",
      pads = [1, 1, 1, 1],
      strides = [1, 1]} : (tensor<1x16x180x320xf32>, tensor<16x16x3x3xf32>, tensor<16xf32>) -> tensor<1x16x180x320xf32> loc(#loc469)
    %475 = "onnx.Relu"(%474) {onnx_node_name = "Relu_442"} : (tensor<1x16x180x320xf32>) -> tensor<1x16x180x320xf32> loc(#loc470)
    %476 = "onnx.Conv"(%475, %174, %173) {
      auto_pad = "NOTSET",
      dilations = [1, 1],
      group = 1 : si64,
      kernel_shape = [1, 1],
      onnx_node_name = "Conv_443",
      pads = [0, 0, 0, 0],
      strides = [1, 1]} : (tensor<1x16x180x320xf32>, tensor<4x16x1x1xf32>, tensor<4xf32>) -> tensor<1x4x180x320xf32> loc(#loc471)
    %477:2 = "onnx.Split"(%476, %1) {axis = 1 : si64, onnx_node_name = "Split_444_34"} : (tensor<1x4x180x320xf32>, tensor<2xi64>) -> (tensor<1x3x180x320xf32>, tensor<1x1x180x320xf32>) loc(#loc472)
    %478 = "onnx.Clip"(%477#1, %69, %75) {onnx_node_name = "Clip_447_13"} : (tensor<1x1x180x320xf32>, tensor<f32>, tensor<f32>) -> tensor<1x1x180x320xf32> loc(#loc473)
    %479 = "onnx.Add"(%477#0, %177) {onnx_node_name = "Add_445"} : (tensor<1x3x180x320xf32>, tensor<1x3x180x320xf32>) -> tensor<1x3x180x320xf32> loc(#loc474)
    %480 = "onnx.Clip"(%479, %69, %75) {onnx_node_name = "Clip_446_1"} : (tensor<1x3x180x320xf32>, tensor<f32>, tensor<f32>) -> tensor<1x3x180x320xf32> loc(#loc475)
    return %478, %468, %450, %431, %409, %480 : tensor<1x1x180x320xf32>, tensor<1x16x90x160xf32>, tensor<1x20x45x80xf32>, tensor<1x40x23x40xf32>, tensor<1x64x12x20xf32>, tensor<1x3x180x320xf32> loc(#loc)
  } loc(#loc)
  "onnx.EntryPoint"() {func = @main_graph} : () -> () loc(#loc)
} loc(#loc)
#loc1 = loc("Initializer_1001")
#loc2 = loc("Initializer_1002")
#loc3 = loc("Initializer_1004")
#loc4 = loc("Initializer_1005")
#loc5 = loc("Initializer_1007")
#loc6 = loc("Initializer_1008")
#loc7 = loc("Initializer_1010")
#loc8 = loc("Initializer_1011")
#loc9 = loc("Initializer_1013")
#loc10 = loc("Initializer_1014")
#loc11 = loc("Initializer_1016")
#loc12 = loc("Initializer_1017")
#loc13 = loc("Initializer_1019")
#loc14 = loc("Initializer_1020")
#loc15 = loc("Initializer_1022")
#loc16 = loc("Initializer_1023")
#loc17 = loc("Initializer_1025")
#loc18 = loc("Initializer_1026")
#loc19 = loc("Initializer_1028")
#loc20 = loc("Initializer_1029")
#loc21 = loc("Initializer_1031")
#loc22 = loc("Initializer_1032")
#loc23 = loc("Initializer_1034")
#loc24 = loc("Initializer_1035")
#loc25 = loc("Initializer_1037")
#loc26 = loc("Initializer_1038")
#loc27 = loc("Initializer_1040")
#loc28 = loc("Initializer_1041")
#loc29 = loc("Initializer_1043")
#loc30 = loc("Initializer_1044")
#loc31 = loc("Initializer_1046")
#loc32 = loc("Initializer_1047")
#loc33 = loc("Initializer_1049")
#loc34 = loc("Initializer_1050")
#loc35 = loc("Initializer_1052")
#loc36 = loc("Initializer_1053")
#loc37 = loc("Initializer_1055")
#loc38 = loc("Initializer_1056")
#loc39 = loc("Initializer_1058")
#loc40 = loc("Initializer_1059")
#loc41 = loc("Initializer_1061")
#loc42 = loc("Initializer_1062")
#loc43 = loc("Initializer_1064")
#loc44 = loc("Initializer_1065")
#loc45 = loc("Initializer_1067")
#loc46 = loc("Initializer_1068")
#loc47 = loc("Initializer_1070")
#loc48 = loc("Initializer_1071")
#loc49 = loc("Initializer_1073")
#loc50 = loc("Initializer_1074")
#loc51 = loc("Initializer_1076")
#loc52 = loc("Initializer_1077")
#loc53 = loc("Initializer_1079")
#loc54 = loc("Initializer_1080")
#loc55 = loc("Initializer_1082")
#loc56 = loc("Initializer_1083")
#loc57 = loc("Initializer_1086")
#loc58 = loc("Initializer_1090")
#loc59 = loc("Initializer_386")
#loc60 = loc("Initializer_388")
#loc61 = loc("Initializer_389")
#loc62 = loc("Initializer_400")
#loc63 = loc("Initializer_752")
#loc64 = loc("Initializer_755")
#loc65 = loc("Initializer_763")
#loc66 = loc("Initializer_809")
#loc67 = loc("Initializer_845")
#loc68 = loc("Initializer_847")
#loc69 = loc("Initializer_890")
#loc70 = loc("Initializer_929")
#loc71 = loc("Initializer_930")
#loc72 = loc("Initializer_932")
#loc73 = loc("Initializer_933")
#loc74 = loc("Initializer_935")
#loc75 = loc("Initializer_936")
#loc76 = loc("Initializer_938")
#loc77 = loc("Initializer_939")
#loc78 = loc("Initializer_941")
#loc79 = loc("Initializer_942")
#loc80 = loc("Initializer_944")
#loc81 = loc("Initializer_945")
#loc82 = loc("Initializer_947")
#loc83 = loc("Initializer_948")
#loc84 = loc("Initializer_950")
#loc85 = loc("Initializer_951")
#loc86 = loc("Initializer_953")
#loc87 = loc("Initializer_954")
#loc88 = loc("Initializer_956")
#loc89 = loc("Initializer_957")
#loc90 = loc("Initializer_959")
#loc91 = loc("Initializer_960")
#loc92 = loc("Initializer_962")
#loc93 = loc("Initializer_963")
#loc94 = loc("Initializer_965")
#loc95 = loc("Initializer_966")
#loc96 = loc("Initializer_968")
#loc97 = loc("Initializer_969")
#loc98 = loc("Initializer_971")
#loc99 = loc("Initializer_972")
#loc100 = loc("Initializer_974")
#loc101 = loc("Initializer_975")
#loc102 = loc("Initializer_977")
#loc103 = loc("Initializer_978")
#loc104 = loc("Initializer_980")
#loc105 = loc("Initializer_981")
#loc106 = loc("Initializer_983")
#loc107 = loc("Initializer_984")
#loc108 = loc("Initializer_986")
#loc109 = loc("Initializer_987")
#loc110 = loc("Initializer_989")
#loc111 = loc("Initializer_990")
#loc112 = loc("Initializer_992")
#loc113 = loc("Initializer_993")
#loc114 = loc("Initializer_995")
#loc115 = loc("Initializer_996")
#loc116 = loc("Initializer_998")
#loc117 = loc("Initializer_999")
#loc118 = loc("Initializer_aspp.aspp2.1.weight")
#loc119 = loc("Initializer_backbone.features.11.block.2.fc1.bias")
#loc120 = loc("Initializer_backbone.features.11.block.2.fc1.weight")
#loc121 = loc("Initializer_backbone.features.11.block.2.fc2.bias")
#loc122 = loc("Initializer_backbone.features.11.block.2.fc2.weight")
#loc123 = loc("Initializer_backbone.features.12.block.2.fc1.bias")
#loc124 = loc("Initializer_backbone.features.12.block.2.fc1.weight")
#loc125 = loc("Initializer_backbone.features.12.block.2.fc2.bias")
#loc126 = loc("Initializer_backbone.features.12.block.2.fc2.weight")
#loc127 = loc("Initializer_backbone.features.13.block.2.fc1.bias")
#loc128 = loc("Initializer_backbone.features.13.block.2.fc1.weight")
#loc129 = loc("Initializer_backbone.features.13.block.2.fc2.bias")
#loc130 = loc("Initializer_backbone.features.13.block.2.fc2.weight")
#loc131 = loc("Initializer_backbone.features.14.block.2.fc1.bias")
#loc132 = loc("Initializer_backbone.features.14.block.2.fc1.weight")
#loc133 = loc("Initializer_backbone.features.14.block.2.fc2.bias")
#loc134 = loc("Initializer_backbone.features.14.block.2.fc2.weight")
#loc135 = loc("Initializer_backbone.features.15.block.2.fc1.bias")
#loc136 = loc("Initializer_backbone.features.15.block.2.fc1.weight")
#loc137 = loc("Initializer_backbone.features.15.block.2.fc2.bias")
#loc138 = loc("Initializer_backbone.features.15.block.2.fc2.weight")
#loc139 = loc("Initializer_backbone.features.4.block.2.fc1.bias")
#loc140 = loc("Initializer_backbone.features.4.block.2.fc1.weight")
#loc141 = loc("Initializer_backbone.features.4.block.2.fc2.bias")
#loc142 = loc("Initializer_backbone.features.4.block.2.fc2.weight")
#loc143 = loc("Initializer_backbone.features.5.block.2.fc1.bias")
#loc144 = loc("Initializer_backbone.features.5.block.2.fc1.weight")
#loc145 = loc("Initializer_backbone.features.5.block.2.fc2.bias")
#loc146 = loc("Initializer_backbone.features.5.block.2.fc2.weight")
#loc147 = loc("Initializer_backbone.features.6.block.2.fc1.bias")
#loc148 = loc("Initializer_backbone.features.6.block.2.fc1.weight")
#loc149 = loc("Initializer_backbone.features.6.block.2.fc2.bias")
#loc150 = loc("Initializer_backbone.features.6.block.2.fc2.weight")
#loc151 = loc("Initializer_decoder.decode1.gru.hh.0.bias")
#loc152 = loc("Initializer_decoder.decode1.gru.hh.0.weight")
#loc153 = loc("Initializer_decoder.decode1.gru.ih.0.bias")
#loc154 = loc("Initializer_decoder.decode1.gru.ih.0.weight")
#loc155 = loc("Initializer_decoder.decode2.gru.hh.0.bias")
#loc156 = loc("Initializer_decoder.decode2.gru.hh.0.weight")
#loc157 = loc("Initializer_decoder.decode2.gru.ih.0.bias")
#loc158 = loc("Initializer_decoder.decode2.gru.ih.0.weight")
#loc159 = loc("Initializer_decoder.decode3.gru.hh.0.bias")
#loc160 = loc("Initializer_decoder.decode3.gru.hh.0.weight")
#loc161 = loc("Initializer_decoder.decode3.gru.ih.0.bias")
#loc162 = loc("Initializer_decoder.decode3.gru.ih.0.weight")
#loc163 = loc("Initializer_decoder.decode4.gru.hh.0.bias")
#loc164 = loc("Initializer_decoder.decode4.gru.hh.0.weight")
#loc165 = loc("Initializer_decoder.decode4.gru.ih.0.bias")
#loc166 = loc("Initializer_decoder.decode4.gru.ih.0.weight")
#loc167 = loc("Initializer_project_mat.conv.bias")
#loc168 = loc("Initializer_project_mat.conv.weight")
#loc169 = loc("Div_2")
#loc170 = loc("Slice_7")
#loc171 = loc("Transpose_8")
#loc172 = loc("Sub_14")
#loc173 = loc("Initializer_398")
#loc174 = loc("Div_16")
#loc175 = loc("Conv_17")
#loc176 = loc("Add_19")
#loc177 = loc("Clip_22")
#loc178 = loc("Div_24")
#loc179 = loc("Mul_25")
#loc180 = loc("Conv_26")
#loc181 = loc("Relu_27")
#loc182 = loc("Conv_28")
#loc183 = loc("Add_29")
#loc184 = loc("Conv_30")
#loc185 = loc("Relu_31")
#loc186 = loc("Conv_32")
#loc187 = loc("Relu_33")
#loc188 = loc("Conv_34")
#loc189 = loc("Conv_35")
#loc190 = loc("Relu_36")
#loc191 = loc("Conv_37")
#loc192 = loc("Relu_38")
#loc193 = loc("Conv_39")
#loc194 = loc("Add_40")
#loc195 = loc("Conv_41")
#loc196 = loc("Relu_42")
#loc197 = loc("Conv_43")
#loc198 = loc("Relu_44")
#loc199 = loc("GlobalAveragePool_45")
#loc200 = loc("Conv_46")
#loc201 = loc("Relu_47")
#loc202 = loc("Conv_48")
#loc203 = loc("Add_50")
#loc204 = loc("Clip_53")
#loc205 = loc("Div_55")
#loc206 = loc("Mul_56")
#loc207 = loc("Conv_57")
#loc208 = loc("Conv_58")
#loc209 = loc("Relu_59")
#loc210 = loc("Conv_60")
#loc211 = loc("Relu_61")
#loc212 = loc("GlobalAveragePool_62")
#loc213 = loc("Conv_63")
#loc214 = loc("Relu_64")
#loc215 = loc("Conv_65")
#loc216 = loc("Add_67")
#loc217 = loc("Clip_70")
#loc218 = loc("Div_72")
#loc219 = loc("Mul_73")
#loc220 = loc("Conv_74")
#loc221 = loc("Add_75")
#loc222 = loc("Conv_76")
#loc223 = loc("Relu_77")
#loc224 = loc("Conv_78")
#loc225 = loc("Relu_79")
#loc226 = loc("GlobalAveragePool_80")
#loc227 = loc("Conv_81")
#loc228 = loc("Relu_82")
#loc229 = loc("Conv_83")
#loc230 = loc("Add_85")
#loc231 = loc("Clip_88")
#loc232 = loc("Div_90")
#loc233 = loc("Mul_91")
#loc234 = loc("Conv_92")
#loc235 = loc("Add_93")
#loc236 = loc("Conv_94")
#loc237 = loc("Add_96")
#loc238 = loc("Clip_99")
#loc239 = loc("Div_101")
#loc240 = loc("Mul_102")
#loc241 = loc("Conv_103")
#loc242 = loc("Add_105")
#loc243 = loc("Clip_108")
#loc244 = loc("Div_110")
#loc245 = loc("Mul_111")
#loc246 = loc("Conv_112")
#loc247 = loc("Conv_113")
#loc248 = loc("Add_115")
#loc249 = loc("Clip_118")
#loc250 = loc("Div_120")
#loc251 = loc("Mul_121")
#loc252 = loc("Conv_122")
#loc253 = loc("Add_124")
#loc254 = loc("Clip_127")
#loc255 = loc("Div_129")
#loc256 = loc("Mul_130")
#loc257 = loc("Conv_131")
#loc258 = loc("Add_132")
#loc259 = loc("Conv_133")
#loc260 = loc("Add_135")
#loc261 = loc("Clip_138")
#loc262 = loc("Div_140")
#loc263 = loc("Mul_141")
#loc264 = loc("Conv_142")
#loc265 = loc("Add_144")
#loc266 = loc("Clip_147")
#loc267 = loc("Div_149")
#loc268 = loc("Mul_150")
#loc269 = loc("Conv_151")
#loc270 = loc("Add_152")
#loc271 = loc("Conv_153")
#loc272 = loc("Add_155")
#loc273 = loc("Clip_158")
#loc274 = loc("Div_160")
#loc275 = loc("Mul_161")
#loc276 = loc("Conv_162")
#loc277 = loc("Add_164")
#loc278 = loc("Clip_167")
#loc279 = loc("Div_169")
#loc280 = loc("Mul_170")
#loc281 = loc("Conv_171")
#loc282 = loc("Add_172")
#loc283 = loc("Conv_173")
#loc284 = loc("Add_175")
#loc285 = loc("Clip_178")
#loc286 = loc("Div_180")
#loc287 = loc("Mul_181")
#loc288 = loc("Conv_182")
#loc289 = loc("Add_184")
#loc290 = loc("Clip_187")
#loc291 = loc("Div_189")
#loc292 = loc("Mul_190")
#loc293 = loc("GlobalAveragePool_191")
#loc294 = loc("Conv_192")
#loc295 = loc("Relu_193")
#loc296 = loc("Conv_194")
#loc297 = loc("Add_196")
#loc298 = loc("Clip_199")
#loc299 = loc("Div_201")
#loc300 = loc("Mul_202")
#loc301 = loc("Conv_203")
#loc302 = loc("Conv_204")
#loc303 = loc("Add_206")
#loc304 = loc("Clip_209")
#loc305 = loc("Div_211")
#loc306 = loc("Mul_212")
#loc307 = loc("Conv_213")
#loc308 = loc("Add_215")
#loc309 = loc("Clip_218")
#loc310 = loc("Div_220")
#loc311 = loc("Mul_221")
#loc312 = loc("GlobalAveragePool_222")
#loc313 = loc("Conv_223")
#loc314 = loc("Relu_224")
#loc315 = loc("Conv_225")
#loc316 = loc("Add_227")
#loc317 = loc("Clip_230")
#loc318 = loc("Div_232")
#loc319 = loc("Mul_233")
#loc320 = loc("Conv_234")
#loc321 = loc("Add_235")
#loc322 = loc("Conv_236")
#loc323 = loc("Add_238")
#loc324 = loc("Clip_241")
#loc325 = loc("Div_243")
#loc326 = loc("Mul_244")
#loc327 = loc("Conv_245")
#loc328 = loc("Add_247")
#loc329 = loc("Clip_250")
#loc330 = loc("Div_252")
#loc331 = loc("Mul_253")
#loc332 = loc("GlobalAveragePool_254")
#loc333 = loc("Conv_255")
#loc334 = loc("Relu_256")
#loc335 = loc("Conv_257")
#loc336 = loc("Add_259")
#loc337 = loc("Clip_262")
#loc338 = loc("Div_264")
#loc339 = loc("Mul_265")
#loc340 = loc("Conv_266")
#loc341 = loc("Conv_267")
#loc342 = loc("Add_269")
#loc343 = loc("Clip_272")
#loc344 = loc("Div_274")
#loc345 = loc("Mul_275")
#loc346 = loc("Conv_276")
#loc347 = loc("Add_278")
#loc348 = loc("Clip_281")
#loc349 = loc("Div_283")
#loc350 = loc("Mul_284")
#loc351 = loc("GlobalAveragePool_285")
#loc352 = loc("Conv_286")
#loc353 = loc("Relu_287")
#loc354 = loc("Conv_288")
#loc355 = loc("Add_290")
#loc356 = loc("Clip_293")
#loc357 = loc("Div_295")
#loc358 = loc("Mul_296")
#loc359 = loc("Conv_297")
#loc360 = loc("Add_298")
#loc361 = loc("Conv_299")
#loc362 = loc("Add_301")
#loc363 = loc("Clip_304")
#loc364 = loc("Div_306")
#loc365 = loc("Mul_307")
#loc366 = loc("Conv_308")
#loc367 = loc("Add_310")
#loc368 = loc("Clip_313")
#loc369 = loc("Div_315")
#loc370 = loc("Mul_316")
#loc371 = loc("GlobalAveragePool_317")
#loc372 = loc("Conv_318")
#loc373 = loc("Relu_319")
#loc374 = loc("Conv_320")
#loc375 = loc("Add_322")
#loc376 = loc("Clip_325")
#loc377 = loc("Div_327")
#loc378 = loc("Mul_328")
#loc379 = loc("Conv_329")
#loc380 = loc("Add_330")
#loc381 = loc("Conv_331")
#loc382 = loc("Add_333")
#loc383 = loc("Clip_336")
#loc384 = loc("Div_338")
#loc385 = loc("Mul_339")
#loc386 = loc("GlobalAveragePool_342")
#loc387 = loc("Conv_343")
#loc388 = loc("Sigmoid_344")
#loc389 = loc("Conv_340")
#loc390 = loc("Relu_341")
#loc391 = loc("Mul_345")
#loc392 = loc("Split_349")
#loc393 = loc("Concat_350")
#loc394 = loc("Conv_351")
#loc395 = loc("Sigmoid_352")
#loc396 = loc("Split_353")
#loc397 = loc("Mul_354")
#loc398 = loc("Concat_355")
#loc399 = loc("Conv_356")
#loc400 = loc("Tanh_357")
#loc401 = loc("Mul_361")
#loc402 = loc("Sub_359")
#loc403 = loc("Mul_360")
#loc404 = loc("Add_362")
#loc405 = loc("Concat_363")
#loc406 = loc("Resize_365")
#loc407 = loc("Slice_371")
#loc408 = loc("AveragePool_346")
#loc409 = loc("AveragePool_347")
#loc410 = loc("AveragePool_348")
#loc411 = loc("Concat_372")
#loc412 = loc("Conv_373")
#loc413 = loc("Relu_374")
#loc414 = loc("Split_375")
#loc415 = loc("Concat_376")
#loc416 = loc("Conv_377")
#loc417 = loc("Sigmoid_378")
#loc418 = loc("Split_379")
#loc419 = loc("Mul_380")
#loc420 = loc("Concat_381")
#loc421 = loc("Conv_382")
#loc422 = loc("Tanh_383")
#loc423 = loc("Mul_387")
#loc424 = loc("Sub_385")
#loc425 = loc("Mul_386")
#loc426 = loc("Add_388")
#loc427 = loc("Concat_389")
#loc428 = loc("Resize_391")
#loc429 = loc("Slice_397")
#loc430 = loc("Concat_398")
#loc431 = loc("Conv_399")
#loc432 = loc("Relu_400")
#loc433 = loc("Split_401")
#loc434 = loc("Concat_402")
#loc435 = loc("Conv_403")
#loc436 = loc("Sigmoid_404")
#loc437 = loc("Split_405")
#loc438 = loc("Mul_406")
#loc439 = loc("Concat_407")
#loc440 = loc("Conv_408")
#loc441 = loc("Tanh_409")
#loc442 = loc("Mul_413")
#loc443 = loc("Sub_411")
#loc444 = loc("Mul_412")
#loc445 = loc("Add_414")
#loc446 = loc("Concat_415")
#loc447 = loc("Resize_417")
#loc448 = loc("Concat_418")
#loc449 = loc("Conv_419")
#loc450 = loc("Relu_420")
#loc451 = loc("Split_421")
#loc452 = loc("Concat_422")
#loc453 = loc("Conv_423")
#loc454 = loc("Sigmoid_424")
#loc455 = loc("Split_425")
#loc456 = loc("Mul_426")
#loc457 = loc("Concat_427")
#loc458 = loc("Conv_428")
#loc459 = loc("Tanh_429")
#loc460 = loc("Mul_433")
#loc461 = loc("Sub_431")
#loc462 = loc("Mul_432")
#loc463 = loc("Add_434")
#loc464 = loc("Concat_435")
#loc465 = loc("Resize_437")
#loc466 = loc("Concat_438")
#loc467 = loc("Conv_439")
#loc468 = loc("Relu_440")
#loc469 = loc("Conv_441")
#loc470 = loc("Relu_442")
#loc471 = loc("Conv_443")
#loc472 = loc("Split_444")
#loc473 = loc("Clip_447")
#loc474 = loc("Add_445")
#loc475 = loc("Clip_446")
#loc476 = loc(fused[#loc172, #loc173])