EDGE-AI-STUDIO: Implementation of Trained Model

Hello,

I've brought this thread to the attention of the model composer team. They will be able to assist further.

Thanks

ki

Sajan,

Regarding: I don't know why I can't do device setup in model composer web app using IP.

Can you confirm that you are able to ssh into your board.  If you can remote into it, then check whether the device and the computer/laptop that you are connecting from are on the same subnet as the device.   For Model Composer to access your device by IP, the device must be on the same subnet as the computer.   
For example, it is likely that if you are connected to a wifi network with your desktop/laptop your IP is not on the same subnet as the device. 

Martin

Hello Martin,

How can i deploy the model. I mean should I need to edit any file in configs or anywhere. 

Can you give me the proper instruction about model deployment. I tried many times but it is not appeared in drop-down option to select.
What should I do after download compiled model to development board.

I builded latest version (sdk 10.0). In the drop down to select model maker in model composer only 2 option available : processor sdk 9.1 and 9.0 . 

Thanks,

Sajan

Unfortunately models compiled for a version of an SDK are not compatible with other SDK version, and at this time SDK 10.x is not supported in Model Composer.   

Since you are running an SDK 10.x image on your device you will need to use the command-line tools.  I am going to forward this to the team that supports these tools. 

Sajan Saji said:

Can you give me the proper instruction about model deployment. I tried many times but it is not appeared in drop-down option to select.

Hello Martin,

Can you please answer this issue. 

Thanks,

Sajan

I have forwarded this thread to the command line tool experts,  In the meantime a starting point would be to review the documentation for ml demo applications that are on the SDK device image and use these for creating your own demo/prototype application.  

https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-am69a/10_01_00/exports/edgeai-docs/common/sample_apps.html#python-c-apps

https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-am69a/10_01_00/exports/edgeai-docs/common/inference_models.html#model-downloader-tool

https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-am69a/10_01_00/exports/edgeai-docs/common/inference_models.html#import-custom-models

I have provided a link to the AM69A SDK documentation,  but the procedure is generally device agnostic. 

A model trained in Model Composer is compatible with all SDKs,   however a trained model that is compiled (in Model Compose or Model Maker) will only be compatible for the compiled-for device and SDK version. 

Hello Martin,

Can you please connect me to command line tool experts asap.

Regards,

Sajan