ExampleInceptionInference - Can't adapt to use a retrained InceptionV3 model and feed it 8 bit grayscale image data it was retrained with.

[PWolly]

Hello Miguel,
I am trying to modify this example to work like a functioning Python equivalent but can't figure out how to get my image type in the right format as evidenced by the exceptions thrown by runner.Run(). The goal is to be able to classify either single image files using the "DecodeJpeg/content" input or image data using the "DecodeJpeg" input. I have included some of the results of my failed efforts in comments in the screenshot. Thanks in advance.

[PWolly]

Got it working when reading .jpg image files using these two lines (borrowed from ImageUtils, CreateTensorFromImageFile())
var contents = File.ReadAllBytes (file);
var tensor = TFTensor.CreateString (contents);
and feeding into the "DecodeJpeg/contents" input.

Now trying to figure out how to reshape decoded 2D image data to feed into "DecodeJpeg".
In the Python code this works;
#Convert the incoming 2D image array into a 3D that the input tensor needs
image_array = cv2.cvtColor(image_data, cv2.COLOR_GRAY2BGR)

@PWolly, Thank you very much. I was experiencing the same problems and your solution worked like a charm.