OpenNCC Frame

# How to write json config to match your model

The json config is an important file to help you run model on NCC.

# nccVideoConvert

This plug-in help you convert video or image data to tensor.

Parameters:

enabled - Flag to enable plug-in

linkIn - Choose input source. (["input"]: external input by USB. ["camera"]: internal input from camera of NCC.)

inputTensor - Shape of input source. Such as: [batch size, number of channels, image height , image width]

inputType - Format of input source. Option: "bgr_planar_u8", "YUV420"

ROI - Region of interest. Such as: [Start_X, Start_Y, End_X, End_Y]

meanValue - No description available

stdValue - No description available

outputTensor - Shape of output tensor. Such as: [batch size, number of channels, image height , image width]

outputType - Format of output tensor.

linkOut - next plug-in

# nccInference

This plug-in help you to inference model on NCC.

Parameters:

enabled - Flag to enable plug-in

linkIn - last plug-in

inputTensor - Shape of input source. Such as: [batch size, number of channels, image height , image width]

inputType - Format of input source. Option: "BGR_PLANAR_FP16"

extInputs1- No description available

extInputs2- No description available

extInputs3- No description available

outputTensor - Shape of output tensor. Such as: [batch size, number of channels, image height , image width]

outputType - Format of output tensor. Option: "CV_FP16"

linkOut - next plug-in

# nccTensorConvert

This plug-in help you to convert format of tensor.

Parameters:

enabled - Flag to enable plug-in

linkIn - last plug-in

inputTensor - Shape of input source. Such as: [batch size, number of channels, image height , image width]

inputType - Format of input source. Option: "CV_FP16"

outputTensor - Shape of output tensor. Such as: [batch size, number of channels, image height , image width]

outputType - Format of output tensor. Option: "CV_FP32"

linkOut - next plug-in or output

In model_zoo, we have provided some reference config for you to run example quickly.

Template is as following:

      "blob":           "<model name>",
      "Function": "This template is used to help you create config with your model.",

      "nccVideoConvert": {
            "enabled":   1,    
            "linkIn"     :  "<source>:[input,camera]",
            "inputTensor": [B,C,H,W],
            "inputType": "<source>:[bgr_planar_u8,YUV420]",    
            "ROI": [Start_X, Start_Y, End_X, End_Y],
            "meanValue": [0,0,0],
            "stdValue": 1,
            "outputTensor": [B,C,H,W],  
            "outputType": "BGR_PLANAR_FP16",      
            "linkOut"     :  "nccInference"       
       },

       "nccInference": {
            "enabled": 1,    
            "linkIn"     :  "nccVideoConvert",
            "inputTensor": [B,C,H,W],  
            "inputType": "BGR_PLANAR_FP16",  
            "extInputs1": [0],  
            "extInputs2": [0],
            "extInputs3": [0],
            "outputTensor": [1,1,200,7],    
            "outputType": "CV_FP16",   
            "linkOut"     :  "nccTensorConvert"     
       },

      "nccTensorConvert": {
            "enabled": 1,
            "linkIn"     :  "nccInference",
            "inputTensor": [1,1,200,7],  
            "inputType": "CV_FP16",  
            "outputTensor": [1,1,200,7],   
            "outputType": "CV_FP32",   
            "linkOut"     :  "output"   
       }

← tensor_filter →