%outputs_18 = "tf.ExpandDims"(%outputs_16, %outputs_0) {device = ""} : (tensor<?x?x3xui8>, tensor<i32>) -> tensor<1x?x?x3xui8>
  %outputs_20 = "tf.ResizeBilinear"(%outputs_18, %outputs) {align_corners = false, device = "", half_pixel_centers = false} : (tensor<1x?x?x3xui8>, tensor<2xi32>) -> tensor<1x224x224x3xf32>

  // b/135640736: MatrixInverse performance issues.
  // b/111271662: MatrixSolve performance issues.
  // https://github.com/tensorflow/tensorflow/pull/31012:
  //    ResizeNearestNeighbor, ResizeBilinear, and ResizeBilinearGrad sometimes
  //    create convolutions too large for CuDNN to handle.
  return node.type_string() == "SelfAdjointEigV2" ||
         node.type_string() == "Svd" || node.type_string() == "Qr" ||

      "RGBToHSV",
      "RandomShuffle",
      "RandomStandardNormal",
      "RandomUniform",
      "RandomUniformInt",
      "ReadVariableOp",
      "ReadVariableXlaSplitND",
      "ResizeBilinear",
      "ResizeBilinearGrad",
      "ResizeNearestNeighbor",
      "ResourceApplyAdaMax",
      "ResourceApplyAdadelta",
      "ResourceApplyAdagrad",
      "ResourceApplyAdagradDA",

}

func.func @resize_with_bilinear(%arg0: tensor<1x100x100x3xf32>, %arg1: tensor<4xi32>) -> tensor<?xf32> {
  %0 = "tf.ResizeBilinear"(%arg0, %arg1) {align_corners = true} : (tensor<1x100x100x3xf32>, tensor<4xi32>) -> tensor<?xf32>
  func.return %0 : tensor<?xf32>
  // CHECK-LABEL: resize_with_bilinear

      TFL_TCresVTEtIsSameAsOp<0, 0>>,
    TFL_OperandHasRank<0, 4>,
    TFL_OperandHasRank<1, 1>,
    SameOperandsAndResultsScale]> {
  let summary = "ResizeBilinear Op";

  let description = [{
    Resize `images` to `size` using bilinear interpolation.
  }];

  let arguments = (ins
    TFL_TensorOf<[F32, TFL_Quint8, QUI8, QI8, QI16]>:$input,

// CHECK-DAG:       %[[SIZE:.*]] = "tf.Const"() <{value = dense<[56, 904]> : tensor<2xi32>}> : () -> tensor<2xi32>
// CHECK:           %[[VAL_2:.*]] = "tf.ResizeBilinear"(%[[VAL_0]], %[[SIZE]]) <{align_corners = true, half_pixel_centers = false}> : (tensor<1x56x624x16xf32>, tensor<2xi32>) -> tensor<1x56x904x16xf32>
// CHECK:           return %[[VAL_2]] : tensor<1x56x904x16xf32>

    OpBuilder<(ins "Value":$tensor, "Value":$shape)>
  ];

  let hasVerifier = 1;

  let hasCanonicalizer = 1;
  let hasFolder = 1;
}

def TF_ResizeBilinearOp : TF_Op<"ResizeBilinear", [Pure]> {
  let summary = "Resize `images` to `size` using bilinear interpolation.";

  let description = [{
Input images can be of different types but output images are always float.
  }];

        is now supported and enabled by default. This adds support for more
        TensorFlow ops including Conv3D, Conv3DBackpropInputV2, AvgPool3D,
        MaxPool3D, ResizeBilinear, and ResizeNearestNeighbor. In addition, the
        TensorFlow-TensorRT python conversion API is exported as
        `tf.experimental.tensorrt.Converter`.