TIDL & TI AI Edge

HelloWorld Demo on Jacinto Edge AI Platform

Software 安装

1. 安装Anaconda
2. 安装Jypyter

   pip3 install pip install jupyterlab
   

3. 安装 TensorflowLite

   $ pip3 install tensorflow 
   $ pip3 install --index-url https://google-coral.github.io/py-repo/ tflite_runtime(Numpy 会有错误直接安装，先不管试试)
   

4. install

   pip install opencv-python==4.2.0.32 --user
   pip install numpy
   

AI EDGE

AI EDGE 是基于TDA4VM 建立的快速进行Deep Learning 评估的云平台，model zoo 有大量的模型可以选择。网页通过python 可以快速部署平台到实际的板子上。　
云平台　https://dev.ti.com/edgeai/
用户指导 https://software-dl.ti.com/jacinto7/esd/edgeai-devkit/00_05_00_02/exports/docs/getting_started.html#overview

Pytorch

精度提升

对浮点转定点会带来精度损失，下面两个方法供参考：
1. Pytorch Jacinto ： https://git.ti.com/cgit/jacinto-ai/pytorch-jacinto-ai-devkit/about/
2. TI https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/07_03_00_07/exports/docs/tidl_j7_02_00_00_07/ti_dl/docs/user_guide_html/md_tidl_fsg_quantization.html

参考

1. https://www.ti.com/tool/SITARA-MACHINE-LEARNING
2. 3 ways machine learning helps companies react quickly to changing circumstances
3. TI Edge Tools: https://www.ti.com/tool/SITARA-MACHINE-LEARNING
4. Deep Learning Inference For Embedded Applications Reference Design
5. Wirte paper Bringing deep learning to embedded systems (Rev. A)
   PDF
6. e2e hello world
7. https://blog.csdn.net/Just_youHG/article/details/104686642
8. github TI https://github.com/TexasInstruments/jacinto-ai-devkit
9. TVM https://zhuanlan.zhihu.com/p/88188955
10. ONNX https://blog.csdn.net/xjm850552586/article/details/114898679
11. tensorflow lite https://tensorflow.google.cn/lite/guide/get_started
12. Edge AIhttps://software-dl.ti.com/jacinto7/esd/edgeai-devkit/00_05_00_02/exports/docs/getting_started.html#overview
13. 神经网络debug的6个实用技巧