Note that, Atlas 200 DK is running on Ubuntu 18.04 ARM OS. You can develop on OS directly (CANN toolkit already installed). In the provided SD card image, there are also two docker containers, which are setup and ready for you to do development. It is for you to share the board between team members without interfering each other. The only restriction of sharing board is that you cannot run multiple atc
commands to convert the model at the same time due to memory limitation.
After the setup is complete, you can begin development on the board following the experiment guide:
https://light-penguin-7f8.notion.site/2021-Introduction-to-Deep-Learning-based-Computer-Vision-and-Edge-Computing-5bf71d8023c94f4e8ef5478d4ec59f54
Video Guide: https://youtu.be/CKsn4NeK2KM
MindStudio
)IMPORTANT NOTE:
MindStudio
can be used to convert the model to.om
format in case you don't know how to useatc
command. It helps to define the input and output nodes including their name and shape. However, you can use https://netron.app/ to find out the input name/shape and fill theatc
command yourself, then run theatc
on 200dk board to convert the model directly as how the Experiment does.
Please find more sample models and their conversion command on https://gitee.com/ascend/samples/tree/master/python/contrib
If you are just converting the model to offline model, it actually does not require Atlas 200DK. You can do it on your own PC, by installing CANN toolkit and better also install MindStudio
IDE. It requires Ubuntu 18.04 (virtual machine is also OK). The following shows you how to setup CANN + MindStudio
environment.
MindStudio
Introduction: https://www.hiascend.com/software/mindstudio