{{indexmenu_n>3}}

====== NPU Model Convert ======

===== Build Virtual Environment =====

The SDK only supports **python3.6** or **python3.8**, here is an example of creating a virtual environment for **python3.8**.

Install python packages.

```shell
$ sudo apt update
$ sudo apt install python3-dev python3-numpy
```

Follow this docs to install [[https://conda.io/projects/conda/en/stable/user-guide/install/linux.html | conda]].

Then create a virtual environment.

```shell
$ conda create -n npu-env python=3.8
$ conda activate npu-env     #activate
$ conda deactivate           #deactivate
```

===== Get Convert Tool =====

Download Tool from [[gh>rockchip-linux/rknn-toolkit2]].

```shell
$ git clone https://github.com/rockchip-linux/rknn-toolkit2.git
$ git checkout 9ad79343fae625f4910242e370035fcbc40cc31a
```

Install dependences and RKNN toolkit2 packages,

```shell
$ cd rknn-toolkit2
$ sudo apt-get install python3 python3-dev python3-pip
$ sudo apt-get install libxslt1-dev zlib1g-dev libglib2.0 libsm6 libgl1-mesa-glx libprotobuf-dev gcc cmake
$ pip3 install -r doc/requirements_cp38-*.txt
$ pip3 install packages/rknn_toolkit2-*-cp38-cp38-linux_x86_64.whl
```

===== Convert Model =====

Converting model has five main steps. Create RKNN object, pre-process config, load model, build model and export RKNN model. Here, take ''yolov5'' ''onnx'' as an example.

Create RKNN object. 

```python
# Create RKNN object
rknn = RKNN(verbose=True) 
```

Pre-process config.

```python
# pre-process config
print('--> Config model')
rknn.config(mean_values=[[0, 0, 0]], std_values=[[255, 255, 255]], target_platform='rk3588')
print('done')
```

  * **mean_values** - The mean of normalization parameter.
  * **std_values** - The variance of normalization parameter.
model input = (image – mean_values) / std_values
  * **target_platform** - Chooses ''rk3588''.

Load model.

```python
# Load ONNX model
print('--> Loading model')
ret = rknn.load_onnx(model='./yolov5.onnx')
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')
```

  * **model** - The path of model.

Load other platform model.

```python
# Load pytorch model
print('--> Loading model')
ret = rknn.load_pytorch(model='./resnet18.pt', input_size_list=[[1, 3, 224, 224]])
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')

# Load tensorflow model
print('--> Loading model')
ret = rknn.load_tensorflow(tf_pb='./ssd_mobilenet_v1_coco_2017_11_17.pb',
                           inputs=['Preprocessor/sub'],
                           outputs=['concat', 'concat_1'],
                           input_size_list=[[300, 300, 3]])
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')

# Load caffe model
print('--> Loading model')
ret = rknn.load_caffe(model='./mobilenet_v2.prototxt', 
                      blobs='./mobilenet_v2.caffemodel')
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')

# Load tensorflow lite model
print('--> Loading model')
ret = rknn.load_tflite(model='./mobilenet_v1.tflite')
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')

# Load darknet model
print('--> Loading model')
ret = rknn.load_darknet(model='./yolov3-tiny.cfg',
                        weight='./yolov3.weights')
if ret != 0:
    print('Load model failed!')
    exit(ret)
print('done')
```

  * **inputs/outputs** - Only use in tensorflow model. It is the name of inputs/outputs.
  * **input_size_list** - The size and channels of input.

Build model

```python
# Build model
print('--> Building model')
ret = rknn.build(do_quantization=True, dataset='./dataset.txt')
if ret != 0:
    print('Build model failed!')
    exit(ret)
print('done')
```

  * **do_quantization** - Quantize model or not.
  * **dataset** - The path of txt file which is written in image path.

Export RKNN model

```python
# Export RKNN model
print('--> Export rknn model')
ret = rknn.export_rknn(export_path='./yolov5_int8.rknn')
if ret != 0:
    print('Export rknn model failed!')
    exit(ret)
print('done')
```

  * **export_path** - The path of rknn model.

All the above codes can be found in ''rknn-toolkit2/examples''. There are all platforms we support now. Choose ''rk3588'' in ''rknn-toolkit2/examples/onnx/yolov5/test.py'' and run the file to convert model.

```diff patch
diff --git a/examples/onnx/yolov5/test.py b/examples/onnx/yolov5/test.py
index a1c9988..f7ce11e 100644
--- a/examples/onnx/yolov5/test.py
+++ b/examples/onnx/yolov5/test.py
@@ -240,7 +240,7 @@ if __name__ == '__main__':

     # pre-process config
     print('--> Config model')
-    rknn.config(mean_values=[[0, 0, 0]], std_values=[[255, 255, 255]])
+    rknn.config(mean_values=[[0, 0, 0]], std_values=[[255, 255, 255]], target_platform='rk3588')
     print('done')
```

Run ''test.py'' to generate rknn model.

```shell
$ python3 test.py
```

<WRAP tip >
In ''test.py'', there are inferring rknn model codes. You can refer it to infer rknn on PC.
</WRAP>

===== See Also =====

For more usage, please refer to the related documents under ''doc''.