Difference between revisions of "AI/RKNN-Toolkit"

Latest revision as of 02:30, 16 March 2020

English

    Home >  AI Development  >  RKNN Toolkit

3) performance evaluation: user can simulate running the model on a PC to get both the total time spent on the model and the time-consuming information of each layer. User can also run the model on the specified hardware platform RK3399Pro/RK1808 by online debugging, and get both the total time of the model running on the hardware and the time-consuming information of each layer.

This chapter mainly explains how to perform model conversion on the RK3399Pro/RK1808 development board. For other function descriptions, please refer to the RKNN-Toolkit User Guide.

Installation preparation

   sudo dnf install -y cmake gcc gcc-c++ protobuf-devel protobuf-compiler lapack-devel
   sudo dnf install -y python3-devel python3-opencv python3-numpy-f2py python3-h5py python3-lmdb  python3-grpcio
   pip3 install scipy-1.2.0-cp36-cp36m-linux_aarch64.whl
   pip3 install onnx-1.4.1-cp36-cp36m-linux_aarch64.whl
   pip3 install tensorflow-1.10.1-cp36-cp36m-linux_aarch64.whl

After installing the above basic package, install the rknn-toolkit wheel package. RKNN wheel package and other Python wheel packages can be downloaded from OneDrive.

Since pip does not have a ready-made aarch64 version of the scipy and onnx wheel packages, we have provided a compiled wheel package. If you want the latest version of the wheel package or find a problem with the pre-compiled wheel package, you can use pip to install it yourself. This will compile and install the wheel package. It will take a long time and you need to wait patiently.

   pip3 install scipy
   pip3 install onnx

If the installation encounters an error, please install the corresponding software package according to the error message.

Model Conversion

API call flow

Example

```
from rknn.api import RKNN
```
```
 
```
```
INPUT_SIZE = 64
```
```
 
```
```
if __name__ == '__main__':
```

    rknn = RKNN()   # Create an RKNN execution object

```
 
```
```
    '''
```

    Configure model input for NPU preprocessing of input data

    channel_mean_value='0 0 0 255', when runing forward inference, the RGB data will be

    converted as follows (R - 0) / 255, (G - 0) / 255, (B - 0) / 255,

    The RKNN model automatically performs the mean and normalization.

    reorder_channel='0 1 2' , used to specify whether to adjust the image channel order,

    set to 0 1 2, means no adjustment according to the input image channel order.

    reorder_channel='2 1 0' , indicates that 0 and 2 channels are exchanged.

    If the input is RGB, it will be adjusted to BGR. If it is BGR will be adjusted to RGB

```
    Image channel order is not adjusted
```
```
    '''
```
```
 
```

    rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')

```
 
```
```
    '''
```
```
    load TensorFlow model
```

    tf_pb='digital_gesture.pb' specify the TensorFlow model to be converted

    inputs specify the input node in the model

    outputs specify the output node in the model

    input_size_list specify the size of the model input

```
    '''
```
```
 
```
```
    print('--> Loading model')
```

    rknn.load_tensorflow(tf_pb='digital_gesture.pb',

                         inputs=['input_x'],

                         outputs=['probability'],

                         input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])

```
    print('done')
```
```
 
```
```
    '''
```
```
    Create a parsing pb model
```

    do_quantization=False do not to be quantified

    Quantization will reduce the size of the model and increase the speed of the operation,

    but there will be loss of precision.

```
    '''
```
```
 
```
```
    print('--> Building model')
```
```
    rknn.build(do_quantization=False)
```
```
    print('done')
```

    rknn.export_rknn('./digital_gesture.rknn')  # Export and save rknn model file

    rknn.release()  # Release RKNN Context

Model Inference

API call flow

Example

```
import numpy as np
```
```
from PIL import Image
```
```
from rknn.api import RKNN
```
```
 
```
```
 
```

# Analyze the output of the model to get the most probable gesture and corresponding probability

```
def get_predict(probability):
```
```
    data = probability[0][0]
```
```
    data = data.tolist()
```
```
    max_prob = max(data)
```

    return data.index(max_prob), max_prob

```
 
```
```
 
```
```
def load_model():
```

    rknn = RKNN()  # Create an RKNN execution object

```
    print('-->loading model')
```

    rknn.load_rknn('./digital_gesture.rknn')  # Load RKNN model

```
    print('loading model done')
```

    print('--> Init runtime environment')

    ret = rknn.init_runtime(host='rk3399pro')  # Initialize the RKNN runtime environment

```
    if ret != 0:
```

        print('Init runtime environment failed')

```
        exit(ret)
```
```
    print('done')
```
```
    return rknn
```
```
 
```
```
 
```
```
def predict(rknn):
```

    im = Image.open("../picture/6_7.jpg")   # load image

    im = im.resize((64, 64), Image.ANTIALIAS)  # Image resize to 64x64

    mat = np.asarray(im.convert('RGB'))    # Convert to RGB format

    outputs = rknn.inference(inputs=[mat])   # Run forward inference and get the inference result

    pred, prob = get_predict(outputs)     # Transform the inference results into visual information

```
    print(prob)
```
```
    print(pred)
```
```
 
```
```
 
```
```
if __name__=="__main__":
```
```
    rknn = load_model()
```
```
    predict(rknn) 
```
```
    rknn.release()
```

Difference between revisions of "AI/RKNN-Toolkit"

Latest revision as of 02:30, 16 March 2020

Contents

Introduction

Installation preparation

Model Conversion

API call flow

Example

Model Inference

API call flow

Example

@@ Line 1: / Line 1: @@
+{{Languages|AI/RKNN-Toolkit}}
+    [[Home | Home]] > [[AI | AI Development]]  > [[AI/RKNN-Toolkit | RKNN Toolkit]]
 === Introduction ===
@@ Line 11: / Line 15: @@
 ) '''performance evaluation''': user can simulate running the model on a PC to get both the total time spent on the model and the time-consuming information of each layer. User can also run the model on the specified hardware platform RK3399Pro/RK1808 by online debugging, and get both the total time of the model running on the hardware and the time-consuming information of each layer.
-This chapter mainly explains how to perform model conversion on the RK3399Pro/RK1808 development board. For other function descriptions, please refer to the RKNN-Toolkit User Guide: "'''RKNN-Toolkit User Guide_V*.pdf'''".
+This chapter mainly explains how to perform model conversion on the RK3399Pro/RK1808 development board. For other function descriptions, please refer to the [https://dl.radxa.com/rockpin10/docs/sw/rknn-toolkit/Rockchip_User_Guide_RKNN_Toolkit_V1.3.0_EN.pdf RKNN-Toolkit User Guide].
 === Installation preparation ===
@@ Line 40: / Line 44: @@
 ==== Example ====
-    <syntaxhighlight lang="Python" line='line'>
+<syntaxhighlight lang="Python" line='line'>
-    from rknn.api import RKNN
+from rknn.api import RKNN
-    INPUT_SIZE = 64
-    if __name__ == '__main__':
-        rknn = RKNN()   # Create an RKNN execution object
-        '''Configure model input for NPU preprocessing of input data
-        channel_mean_value='0 0 0 255', when runing forward inference, the RGB data will be converted as follows
-        (R - 0)/255, (G - 0)/255, (B - 0)/255, The RKNN model automatically performs the mean and normalization
-        reorder_channel=' 0 1 2' , used to specify whether to adjust the image channel order, set to 0 1 2, means no adjustment according to the input image channel order.
-        reorder_channel=' 2 1 0' , indicates that 0 and 2 channels are exchanged. If the input is RGB, it will be adjusted to BGR. If it is BGR will be adjusted to RGB
-        Image channel order is not adjusted'''
-        rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')
-        '''
-        load TensorFlow model
-        tf_pb='digital_gesture.pb' specify the TensorFlow model to be converted
-        inputs specify the input node in the model
-        outputs specify the output node in the model
-        input_size_list specify the size of the model input
-        '''
-        print('--> Loading model')
-        rknn.load_tensorflow(tf_pb='digital_gesture.pb',
-                             inputs=['input_x'],
-                             outputs=['probability'],
-                             input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])
-        print('done')
-        '''
+INPUT_SIZE = 64
-        Create a parsing pb model
-        do_quantization=False do not to be quantified
+if __name__ == '__main__':
-        Quantization will reduce the size of the model and increase the speed of the operation, but there will be loss of precision.
+    rknn = RKNN()   # Create an RKNN execution object
-        '''
+    '''
-        print('--> Building model')
+    Configure model input for NPU preprocessing of input data
-        rknn.build(do_quantization=False)
+    channel_mean_value='0 0 0 255', when runing forward inference, the RGB data will be
-        print('done')
+    converted as follows (R - 0) / 255, (G - 0) / 255, (B - 0) / 255,
-        rknn.export_rknn('./digital_gesture.rknn')  # Export and save rknn model file
+    The RKNN model automatically performs the mean and normalization.
-        rknn.release()  # Release RKNN Context
+    reorder_channel='0 1 2' , used to specify whether to adjust the image channel order,
-    </syntaxhighlight>
+    set to 0 1 2, means no adjustment according to the input image channel order.
+    reorder_channel='2 1 0' , indicates that 0 and 2 channels are exchanged.
+    If the input is RGB, it will be adjusted to BGR. If it is BGR will be adjusted to RGB
+    Image channel order is not adjusted
+    '''
+    rknn.config(channel_mean_value='0 0 0 255', reorder_channel='0 1 2')
+    '''
+    load TensorFlow model
+    tf_pb='digital_gesture.pb' specify the TensorFlow model to be converted
+    inputs specify the input node in the model
+    outputs specify the output node in the model
+    input_size_list specify the size of the model input
+    '''
+    print('--> Loading model')
+    rknn.load_tensorflow(tf_pb='digital_gesture.pb',
+                         inputs=['input_x'],
+                         outputs=['probability'],
+                         input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])
+    print('done')
+    '''
+    Create a parsing pb model
+    do_quantization=False do not to be quantified
+    Quantization will reduce the size of the model and increase the speed of the operation,
+    but there will be loss of precision.
+    '''
+    print('--> Building model')
+    rknn.build(do_quantization=False)
+    print('done')
+    rknn.export_rknn('./digital_gesture.rknn')  # Export and save rknn model file
+    rknn.release()  # Release RKNN Context
+</syntaxhighlight>
 === Model Inference ===
@@ Line 93: / Line 103: @@
 ==== Example ====
-    import numpy as np
+<syntaxhighlight lang="Python" line='line'>
-    from PIL import Image
+import numpy as np
-    from rknn.api import RKNN
+from PIL import Image
+from rknn.api import RKNN
-    # Analyze the output of the model to get the most probable gesture and corresponding probability
-    def get_predict(probability):
+# Analyze the output of the model to get the most probable gesture and corresponding probability
-        data = probability[0][0]
+def get_predict(probability):
-        data = data.tolist()
+    data = probability[0][0]
-        max_prob = max(data)
+    data = data.tolist()
-        return data.index(max_prob), max_prob;
+    max_prob = max(data)
+    return data.index(max_prob), max_prob
-    def load_model():
-        rknn = RKNN()  # Create an RKNN execution object
+def load_model():
-        print('-->loading model')
+    rknn = RKNN()  # Create an RKNN execution object
-        rknn.load_rknn('./digital_gesture.rknn')  # Load RKNN model
+    print('-->loading model')
-        print('loading model done')
+    rknn.load_rknn('./digital_gesture.rknn')  # Load RKNN model
-        print('--> Init runtime environment')
+    print('loading model done')
-        ret = rknn.init_runtime(host='rk3399pro')  # Initialize the RKNN runtime environment
+    print('--> Init runtime environment')
-        if ret != 0:
+    ret = rknn.init_runtime(host='rk3399pro')  # Initialize the RKNN runtime environment
-            print('Init runtime environment failed')
+    if ret != 0:
-            exit(ret)
+        print('Init runtime environment failed')
-        print('done')
+        exit(ret)
-        return rknn
+    print('done')
+    return rknn
-    def predict(rknn):
-        im = Image.open("../picture/6_7.jpg")   # load image
+def predict(rknn):
-        im = im.resize((64, 64),Image.ANTIALIAS)  # Image resize to 64x64
+    im = Image.open("../picture/6_7.jpg")   # load image
-        mat = np.asarray(im.convert('RGB'))    # Convert to RGB format
+    im = im.resize((64, 64), Image.ANTIALIAS)  # Image resize to 64x64
-        outputs = rknn.inference(inputs=[mat])   # Run forward inference and get the inference result
+    mat = np.asarray(im.convert('RGB'))    # Convert to RGB format
-        pred, prob = get_predict(outputs)     # Transform the inference results into visual information
+    outputs = rknn.inference(inputs=[mat])   # Run forward inference and get the inference result
-        print(prob)
+    pred, prob = get_predict(outputs)     # Transform the inference results into visual information
-        print(pred)
+    print(prob)
+    print(pred)
-    if __name__=="__main__":
-        rknn = load_model()
+if __name__=="__main__":
-        predict(rknn)
+    rknn = load_model()
-        rknn.release()
+    predict(rknn)
+    rknn.release()
+</syntaxhighlight>