Skip to content

Latest commit

 

History

History
187 lines (121 loc) · 6.9 KB

Quick_start_with_RT-AK_K210_EN.md

File metadata and controls

187 lines (121 loc) · 6.9 KB
Raw

中文: Chinese

Quick Start for RT-AK with Plugin K210

[TOC]

1. Preparation

Please prepare these important materials as below:

Index Prepare Example
1 Hardware、BSP and Cross compilation tool chain K210 BSP
xpack-riscv-none-embed-gcc-8.3.0-1.2
2 RT-AK RT-AK
3 AI model ./rt_ai_tools/Model/mnist.tflite
4 K210 offical tools NNCase
KFlash

1.1 BSP

1.2 RT-AK

Please clone RT-AK

$ git clone https://github.com/RT-Thread/RT-AK.git edge-ai

1.3 AI model

k210 offical tool supports three kinds of AI model typies： TFLite、Caffe、ONNX

Transform the AI model into kmodel, which is located at RT-AK/rt_ai_tools/Models/mnist.tflite

1.4 k210 offical tool

  1. NNCase：NNCase is located at RT-AK/rt_ai_tools/platforms/plugin_k210/k_tools

    You can also download it form Github

  2. K-Flash

2. Execution steps

2.1 Basic commands

Please run aitools at edge-ai/RT-AK/tools

image-20210616200108220

During the running process of the RT-AK

  1. K210 plug-in will be pulled from github to RT-AK/rt_ai_tools/platforms automatically
  2. AI model will be integrated in the BSP, but not include the application codes of the AI model, an example is presented at the end of the doc
  3. In the PATH of RT-AK/rt_ai_tools/platforms/plugin_k210, <model_name>.kmodel and convert_report.txt will be generated
    • <model_name>.kmodel AI model for K210
    • convert_report.txt log for AI model transformation

image-20210617111819068

image-20210617112301513

# Basic commands
python aitools.py --project=<your_project_path> --model=<your_model_path> --model_name=<your_model_name> --platform=k210 --clear

# Examples
$ D:\Project\edge-ai\RT-AK\rt_ai_tools>python aitools.py --project=D:\Project\K210_Demo\k210_rthread_bsp --model=.\Models\mnist.tflite --model_name=mnist --platform=k210 --embed_gcc=D:\Project\k210_third_tools\xpack-riscv-none-embed-gcc-8.3.0-1.2\bin --dataset=.\platforms\plugin_k210\datasets\mnist_datasets

An successful demo for RT-AK：

2.2 Other additional notes

# no model quantization, --inference_type
$ python aitools.py --project=<your_project_path> --model=<your_model_path> --platform=k210  --inference_type=float

# no model quantization, set path for the cross compilation tool chain
$ python aitools.py --project=<your_project_path> --model=<your_model_path> --platform=k210 --embed_gcc=<your_RISCV-GNU-Compiler_path> --inference_type=float

# model quantization with uint8 type, accelerating computation with KPU, dataset format is image
$ python aitools.py --project=<your_project_path> --model=<your_model_path> --platform=k210 --embed_gcc=<your_RISCV-GNU-Compiler_path> --dataset=<your_val_dataset>

# model quantization with uint8 type, accelerating computation with KPU, dataset format is not image
$ python aitools.py --project=<your_project_path> --model=<your_model_path> --platform=k210 --embed_gcc=<your_RISCV-GNU-Compiler_path> --dataset=<your_val_dataset> --dataset_format=raw

# example
$ python aitools.py --project="D:\Project\k210_val" --model="./Models/facelandmark.tflite" --model_name=facelandmark --platform=k210 --embed_gcc="D:\Project\k210_third_tools\xpack-riscv-none-embed-gcc-8.3.0-1.2\bin" --dataset="./platforms/plugin_k210/datasets/images"

2.3 Input parameters of RT-AK

Parameter Description
--project OS+BSP project folder，should be assigned by user
--model AI model PATH. Default is ./Models/keras_mnist.h5
--model_name New AI model name after model transformation. Default is network
--platform Target platform, such as: stm32, k210，Default is example
--embed_gcc PATH for cross compilation tool chain
--dataset Dataset for AI model quantization

3. Compilation & Download

Compilation:

scons -j 6

rtthread.bin will be generated in the folder

Download：

By using the K-Flash, you can download the rtthread.bin into the test board

4. Explanation for embed application project

We have provide a open source test project, which you can download from here

4.1 Work flow of the RT-Thread RTOS

System initialization：

  • System clock initialization

Loading and running the AI model：

  • Registering the AI model
  • Finding AI model
  • Initializing AI model
  • Inferring AI model
  • Getting results

4.2 Core codes

// main.c
/* Set CPU clock */
sysctl_clock_enable(SYSCTL_CLOCK_AI);  // System clock initialization
...

/* AI modol inference */
mymodel = rt_ai_find(MY_MODEL_NAME);  // find AI model
if (rt_ai_init(mymodel, (rt_ai_buffer_t *)IMG9_CHW) != 0)  // Initializing AI model
...
if(rt_ai_run(mymodel, ai_done, NULL) != 0)    // Inferring
...
output = (float *)rt_ai_output(mymodel, 0);  // Getting results

/* Getting the high */
for(int i = 0; i < 10 ; i++)
{
    // printf("pred: %d, scores: %.2f%%\n", i, output[i]*100);
    if(output[i] > scores  && output[i] > 0.2)
    {
        prediction = i;
        scores = output[i];
    }
}

How to change the input image:

In applications folder, you can only modify the line 18 and line 51

The MNIST demo can be download from Github

4.3 Results of MNIST demo

image-20210616170010919