Using Onnx trained on Keras with PreprocessInput

[bbday]

I have issue when i using onnx model trained with keras and preprocess_input, is ResNet50 model,
if i training without preprocess_input on keras it work on ml.net but when i use preprocess_input will not work, according documentation preprocess_input will process image with caffe style that mean:

The images are converted from RGB to BGR, then each color channel is zero-centered with respect to the ImageNet dataset, without scaling.

this work without preprocess_input

var dataschema = mlContext.Data.LoadFromEnumerable(new List<ImageNetData>());
            var transformer =
                mlContext.Transforms.ExtractPixels(outputColumnName: "input_1", nameof(ImageNetData.BitImage), interleavePixelColors: true)
                .Append(mlContext.Transforms.ApplyOnnxModel(modelFile: path, outputColumnNames: new[] { "fc360" }, inputColumnNames: new[] { "input_1" })) //gpuDeviceId:0
                .Fit(dataschema);

i tried this convert from RGB to BGR but i dont see any parameter for zero-centered color mode

mlContext.Transforms.ExtractPixels(outputColumnName: "input_1", nameof(ImageNetData.BitImage), orderOfExtraction: ImagePixelExtractingEstimator.ColorsOrder.ABGR, interleavePixelColors: true)

Maybe i miss something ???