text-recognition-0012: Input preprocessing from BGR to grayscale

[chixyuan]

Hi, I'm trying to use text-recognition-0012 but there is no model proc sample for that model. How can I transform the BGR input into 1-channel input to accommodate with the model requirement?
https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/intel/text-recognition-0012/README.md

current model_proc:
{
"json_schema_version": "2.0.0",
"input_preproc": [
{
"layer_name": "Placeholder",
"format": "image",
"resize": {
"width": 120,
"height": 32
},
"params": {
"color_space": "GRAYSCALE",
"range": [0.0, 1.0],
"mean": [0.485],
"std": [0.229]
}
}
],
"output_postproc": [
{
"layer_name": "shadow/LSTMLayers/transpose_time_major",
"format": "text",
"converter": "label",
"method": "index",
"params": {
"alphabet": "0123456789abcdefghijklmnopqrstuvwxyz#"
}
}
]
}

gst-launch-1.0 filesrc location=my_video.mp4 ! decodebin ! gvaclassify model=text-recognition-0012.xml model_proc=my_proc_file.json ...

Error:
base_inference plugin initialization failed (3): /home/dlstreamer/dlstreamer/src/monolithic/gst/inference_elements/base/inference_singleton.cpp(181): acquire_inference_instance (): /GstPipeline:pipeline14/GstGvaClassify:gvaclassify4:\n\nFailed to construct OpenVINOImageInference\n\tCheck 'node_shape[channels_index].is_dynamic() || node_shape[channels_index] == values_size' failed at src/core/src/preprocess/preprocess_steps_impl.cpp:28:\nNumber of channels and mean/values size mismatch: Channels = 1, mean/scale = 3, shape = [1,1,32,120], layout = [N,C,H,W]\n", "module": "gstreamer_pipeline"}