radxa-docs · peterwang2050 · Sep 11, 2024 · Sep 11, 2024 · Sep 11, 2024
@@ -10,9 +10,11 @@
 - 板端利用 rknn-toolkit2-lite 的 Python API 板端推理模型
 
 ### PC端模型转换
+
 :::tip
 Radxa 已提供预转换好的 `yolov5s_rk35XX.rknn` 模型，用户可直接参考[板端推理 YOLOv5 ](#板端推理-yolov5)跳过 PC 端模型转换章节
 :::
+
 - 如使用 conda 请先激活 rknn conda 环境
 
   ```bash
@@ -49,41 +51,49 @@ Radxa 已提供预转换好的 `yolov5s_rk35XX.rknn` 模型，用户可直接参
 - 将 yolov5.rknn 模型拷贝到板端
 
 ### 板端推理 YOLOv5
+
 :::tip
 RK3566/3568 芯片用户使用 NPU 前需要在 rsetup overlays 中开启, 具体请参考 [rsetup](../radxa-os/rsetup)
 :::
 
 - （可选）下载 radxa 准备的 yolov5s rknn 模型
-    | 平台 | 下载链接                                                         |
-    | -------- | ------------------------------------------------------------ |
-    | rk3566   | [yolov5s_rk3566.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3566.rknn) |
-    | rk3568   | [yolov5s_rk3568.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3568.rknn) |
-    | rk3588   | [yolov5s_rk3588.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3588.rknn) |
+  | 平台 | 下载链接 |
+  | -------- | ------------------------------------------------------------ |
+  | rk3566 | [yolov5s_rk3566.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3566.rknn) |
+  | rk3568 | [yolov5s_rk3568.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3568.rknn) |
+  | rk3588 | [yolov5s_rk3588.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3588.rknn) |
 
 - 修改 `rknn_model_zoo/py_utils/rknn_executor.py` 代码
-    ```python
-    1 # from rknn.api import RKNN
-    2 try:
-    3     from rknn.api import RKNN
-    4 except:
-    5     from rknnlite.api import RKNNLite as RKNN
-    ...
-    ...
-    18 ret = rknn.init_runtime()
-    ```
-- 修改 `rknn_model_zoo/examples/yolov5/python` 代码
-    ```python
-    262 outputs = model.run([np.expand_dims(input_data, 0)])
-    ```
+
+  请根据[板端安装 RKNN Model Zoo](./rknn_install#可选-板端安装-rknn-model-zoo) 配置 RKNN Model Zoo 代码仓库
+
+  ```python
+  1 # from rknn.api import RKNN
+  2 try:
+  3     from rknn.api import RKNN
+  4 except:
+  5     from rknnlite.api import RKNNLite as RKNN
+  ...
+  ...
+  18 ret = rknn.init_runtime()
+  ```
+
+- 修改 `rknn_model_zoo/examples/yolov5/python/yolov5.py` 代码
+  ```python
+  262 outputs = model.run([np.expand_dims(input_data, 0)])
+  ```
+- 安装依赖环境
+  ```bash
+  pip3 install opencv-python-headless
+  ```
 - 运行 yolov5 示例代码
-  请根据[板端安装 RKNN Model Zoo](./rknn_install#可选-板端安装-rknn-model-zoo)配置 RKNN Model Zoo 代码仓库
 
   ```bash
   cd rknn_model_zoo/examples/yolov5/python
   python3 yolov5.py --model_path <your model path> --img_save
   ```
-  如你使用的是自己转换的模型需从 PC 端拷贝到板端，并用 --model_path 参数指定模型路径
 
+  如你使用的是自己转换的模型需从 PC 端拷贝到板端，并用 --model_path 参数指定模型路径
 
   ```bash
   rock@radxa-zero3:~/rknn_model_zoo/examples/yolov5/python$ python3 yolov5.py --model_path ./yolov5s_rk3566.rknn --target rk3566 --img_save

@@ -1,20 +1,50 @@
 ---
 sidebar_position: 1
 ---
+
 # 周易Z2 AIPU
 
 “周易” AIPU 是由安谋中国针对深度学习而自主研发的创新性 AI 专用处理器，它采用了创新性的架构设计，提供完整的硬件和软件生态，并且具有 PPA 最佳平衡。
 安谋中国还为“周易” AIPU 的客户提供很多工具来帮助他们进行开发，包括仿真器、编译器和调试器等进行数据的采集、分析。
 “周易” AIPU 也支持业界主流的 AI 规模框架，包括 TensorFlow、ONNX 等，未来也将支持更多不同的扩展框架。
 
 “周易” Z2 AIPU 将主要面向中高端安防、智能座舱和 ADAS、边缘服务器等应用场景。
+
+## 快速例子
+
+radxa 提供一个开箱即用的目标分类例子，旨在用户可以直接在 sirider s1 使用 AIPU 推理 resnet50 模型，免去复杂的模型编译和执行代码编译，
+这对想快速使用 AIPU 而不想从头编译模型的用户是最佳的选择，如您对完整工作流程感兴趣可以参考 [周易 Z2 AIPU 使用教程](zhouyi_npu#周易-z2-aipu-使用教程)
+
+- 克隆仓库代码
+  ```bash
+  git clone https://github.com/zifeng-radxa/siriders1_NPU_example.git
+  ```
+- 安装依赖
+  ```bash
+  cd siriders1_NPU_example
+  pip3 install -r requirements.txt
+  ```
+- 生成用于模型输入的文件
+
+  ```bash
+  python3 input_gen.py --img_path <your_image_path>
+  ```
+
+- 模型推理
+  ```bash
+  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$(pwd)/libs
+  ./aipu_test aipu_mlperf_resnet50.bin input_3_224_224.bin
+  ```
+
 ## 周易 Z2 AIPU 使用教程
 
 ### x86 PC 端安装周易 AIPU SDK
+
 周易 SDK 是一个全栈平台，可为用户提供快速上市的开发和部署能力。
 ![image](https://user-images.githubusercontent.com/85479712/198521602-49e13a31-bb49-424f-b782-5108274d63c3.png)
 
 - 准备一个 python3.8 的环境
+
   - （可选）安装 [Anaconda](https://www.anaconda.com/)
 
     如果系统中没有安装 Python 3.8（必要版本），或者同时有多个版本的 Python 环境，建议使用 [Anaconda](https://www.anaconda.com/) 创建新的 Python 3.8 环境
@@ -40,35 +70,39 @@ sidebar_position: 1
       conda activate aipu
       ```
 
-    - 退出环境 
+    - 退出环境
       ```bash
       conda deactivate
       ```
-      
+
 - 在[瑞莎下载站](https://dl.radxa.com/sirider/s1/)下载周易 Z2 SDK 安装包后解压安装
-    ```bash
-    tar -xvf Zhouyi_Z2.tar.gz
-    cd Zhouyi_Z2 && bash +x SETUP.SH
-    ```
+  ```bash
+  tar -xvf Zhouyi_Z2.tar.gz
+  cd Zhouyi_Z2 && bash +x SETUP.SH
+  ```
 - 安装后得到的完整 SDK 文件如下
 
   - `AI610-SDK-r1p3-AIoT` : ARM ZhouYi Z2 工具包
 
   - `siengine` : siengine 提供的 ARM ZhouYi Z2 模型编译(nn-compiler-user-case-example)及板子部署(nn-runtime-user-case-example)的 demos
 
 - 配置 nn-compiler 环境
-    ```bash
-    cd AI610-SDK-r1p3-AIoT/AI610-SDK-r1p3-00eac0/Out-Of-Box/out-of-box-nn-compiler
-    pip3 install -r lib_dependency.txt
-    ```
+
+  ```bash
+  cd AI610-SDK-r1p3-AIoT/AI610-SDK-r1p3-00eac0/Out-Of-Box/out-of-box-nn-compiler
+  pip3 install -r lib_dependency.txt
+  ```
+
   因为此 SDK 不包含模拟功能, 故安装过程会出现安装 AIPUSimProfiler 的报错，可以忽略
-  
+
   若使用 venv 的用户请在 env_setup.sh 中 pip3 install 部分去掉 --user 选项
-    ```bash
-    source env_setup.sh
-    ```
+
+  ```bash
+  source env_setup.sh
+  ```
 
 ### x86 PC 端模型转换
+
 nn-compiler 可以将 TensorFlow、ONNX 等框架模型转换成可以在周易 AIPU 进行硬件加速推理的模型文件
 :::tip
 此案例中将介绍开箱即用案例：resnet50 目标分类
@@ -89,29 +123,36 @@ nn-compiler 可以将 TensorFlow、ONNX 等框架模型转换成可以在周易
   python3 generate_calibration_data.py
   ```
 - 生成用于模型推理的照片文件
+
   ```bash
   python3 generate_input_binary.py
   ```
+
   文件在 ./resnet50/input_3_224_224.bin
 
-- (可选) 配置 build.cfg （开箱即用案例已提供） 
+- (可选) 配置 build.cfg （开箱即用案例已提供）
   ```bash
   vim ./resnet50/build.cfg
   ```
 - 生成 aipu 模型
+
   ```bash
   cd ./restnet50
   aipubuild build.cfg
   ```
+
   在 ./restnet50 中得到 aipu_mlperf_resnet50.bin
 
   :::tip
   如果找不到 `aipubuild` 命令，可尝试 `export PATH=$PATH:/root/.local/bin`
   :::
 
 ### 板端使用周易 Z2 推理 AIPU 模型
+
 在使用周易 Z2 AIPU 推理前需要在 x86 主机进行交叉编译生成可执行文件 `aiputest`，然后拷贝到 Sirider S1 中执行
+
 #### 在 x86 PC 端交叉编译二进制可执行文件
+
 - 安装 [gcc-linaro-7.5.0-2019.12-x86_64_aarch64-linux-gnu](https://releases.linaro.org/components/toolchain/binaries/latest-7/aarch64-linux-gnu/) 交叉编译工具链
   ```bash
   tar -xvf gcc-linaro-7.5.0-2019.12-x86_64_aarch64-linux-gnu.tar
@@ -120,30 +161,37 @@ nn-compiler 可以将 TensorFlow、ONNX 等框架模型转换成可以在周易
 - 编译 aiputest
 
   - 修改 UMDSRC 变量
+
   ```bash
-  cd siengine/nn-runtime-user-case-example 
+  cd siengine/nn-runtime-user-case-example
   vim CMakeLists.txt
   #set(UMDSRC "${CMAKE_SOURCE_DIR}/../AI610-SDK-${AIPU_VERSION}-00eac0/AI610-SDK-1012-${AIPU_VERSION}-eac0/Linux-driver/driver/umd")
   set(UMDSRC "${CMAKE_SOURCE_DIR}/../../AI610-SDK-${AIPU_VERSION}-AIoT/AI610-SDK-r1p3-00eac0/AI610-SDK-1012-${AIPU_VERSION}-eac0/Linux-driver/driver/umd")
   ```
+
   - 交叉编译
+
   ```bash
   mkdir build && cd build
   cmake -DCMAKE_BUILD_TYPE=Release ..
   make
   ```
+
   编译生成的文件在 `siengine/nn-runtime-user-case-example/out/linux/aipu_test`
 
 #### 在 Sirider S1 进行板端推理
+
 - 将生成的 `aipu_mlperf_resnet50.bin` 模型文件，`input_3_224_224.bin` 照片文件，`aipu_test` 可执行文件，`out/linux/libs` 动态库文件夹复制到 Sirider S1 中
 - 执行 aipu_test
+
   ```bash
   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<your libs path>
   ./aipu_test aipu_mlperf_resnet50.bin input_3_224_224.bin
   ```
+
   ```bash
-  (aiot-focal_overlayfs)root@linux:~/ssd# ./aipu_test aipu_mlperf_resnet50.bin input_3_224_224.bin 
-  usage: ./aipu_test aipu.bin input0.bin 
+  (aiot-focal_overlayfs)root@linux:~/ssd# ./aipu_test aipu_mlperf_resnet50.bin input_3_224_224.bin
+  usage: ./aipu_test aipu.bin input0.bin
   aipu_init_context success
   aipu_load_graph_helper success: aipu_mlperf_resnet50.bin
   aipu_create_job success
@@ -171,13 +219,15 @@ nn-compiler 可以将 TensorFlow、ONNX 等框架模型转换成可以在周易
   aipu_unload_graph success
   aipu_deinit_ctx success
   ```
+
   两次的推理总时间
+
   ```bash
   real	0m0.043s
   user	0m0.008s
   sys	0m0.023s
   ```
-  
+
   这里结果仅显示 推理结果的标签值，最大置信度 637 即对应 [imagenet1000](https://gist.github.com/yrevar/942d3a0ac09ec9e5eb3a) 中的 `mailbag, postbag`
 
-  ![input.webp](/img/sirider/s1/aipu_1.webp)
+  ![input.webp](/img/sirider/s1/aipu_1.webp)
@@ -10,9 +10,11 @@ Deploying YOLOv5 with RKNN requires two steps:
 - On the board, use the Python API of `rknn-toolkit2-lite` for on-board model inference.
 
 ### PC Model Conversion
+
 :::tip
 Radxa provides a pre-converted `yolov5s_rk35XX.rknn` model, and users can directly refer to [YOLOv5 On-Board Inference](#yolov5-on-board-inference) to skip the PC model conversion section.
 :::
+
 - If you are using conda, please activate the rknn conda environment first.
 
   ```bash
@@ -49,39 +51,50 @@ Radxa provides a pre-converted `yolov5s_rk35XX.rknn` model, and users can direct
 - Copy the `yolov5.rknn` model to the board.
 
 ### YOLOv5 On-Board Inference
+
 :::tip
 For RK3566/3568 chip users, NPU must be enabled in rsetup overlays before use. Please refer to [rsetup](../radxa-os/rsetup) for details.
 :::
 
 - (Optional) Download the Radxa-provided YOLOv5s RKNN model.
-    | Platform | Download Link                                                         |
-    | -------- | --------------------------------------------------------------------- |
-    | rk3566   | [yolov5s_rk3566.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3566.rknn) |
-    | rk3568   | [yolov5s_rk3568.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3568.rknn) |
-    | rk3588   | [yolov5s_rk3588.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3588.rknn) |
+  | Platform | Download Link |
+  | -------- | --------------------------------------------------------------------- |
+  | rk3566 | [yolov5s_rk3566.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3566.rknn) |
+  | rk3568 | [yolov5s_rk3568.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3568.rknn) |
+  | rk3588 | [yolov5s_rk3588.rknn](https://github.com/zifeng-radxa/rknn_model_zoo/releases/download/yolov5s_rknn/yolov5s_rk3588.rknn) |
 
 - Modify the `rknn_model_zoo/py_utils/rknn_executor.py` code
-    ```python
-    1 # from rknn.api import RKNN
-    2 try:
-    3     from rknn.api import RKNN
-    4 except:
-    5     from rknnlite.api import RKNNLite as RKNN
-    ...
-    ...
-    18 ret = rknn.init_runtime()
-    ```
-- Modify the `rknn_model_zoo/examples/yolov5/python` code
-    ```python
-    262 outputs = model.run([np.expand_dims(input_data, 0)])
-    ```
-- Run the YOLOv5 example code
+
   Please configure the RKNN Model Zoo code repository according to [Install RKNN Model Zoo on the Board](./rknn_install#optional-install-rknn-model-zoo-on-the-board).
 
+  ```python
+  1 # from rknn.api import RKNN
+  2 try:
+  3     from rknn.api import RKNN
+  4 except:
+  5     from rknnlite.api import RKNNLite as RKNN
+  ...
+  ...
+  18 ret = rknn.init_runtime()
+  ```
+
+- Modify the `rknn_model_zoo/examples/yolov5/python/yolov5.py` code
+
+  ```python
+  262 outputs = model.run([np.expand_dims(input_data, 0)])
+  ```
+
+- Install the required environment
+  ```bash
+  pip3 install opencv-python-headless
+  ```
+- Run the YOLOv5 example code
+
   ```bash
   cd rknn_model_zoo/examples/yolov5/python
   python3 yolov5.py --model_path <your model path> --img_save
   ```
+
   If you are using a self-converted model, copy it from the PC to the board, and specify the model path with the `--model_path` parameter.
 
   ```bash
@@ -117,4 +130,4 @@ For RK3566/3568 chip users, NPU must be enabled in rsetup overlays before use. P
 
 - All inference results are saved in `./result`.
 
-![result0](/img/general-tutorial/rknn/result.webp)
+![result0](/img/general-tutorial/rknn/result.webp)