Is it possible to convert a deep neural network to C++ and then deploy it to run on a DSP?

Is it possible to convert a deep neural network to C++ and then deploy it to run on a DSP?

Answer 1

[檮杌]

In a simple answer, yes, you can. C/C++ code generation is supported for many neural network layers. You can refer to the doc below to see networks and layers that are supported for code generation.

https://www.mathworks.com/help/coder/ug/networks-and-layers-supported-for-c-code-generation.html

Also, below is a hands-on example to generate a C/C++ code with a neural network.

https://www.mathworks.com/help/coder/ug/generate-generic-cc-code-for-deep-learning-networks.html

However, more information is required in order to get a precise answer. For example, what "DSP" device are you mentioning about? In other words, what's your target device? How is your network composed of? What is your workflow? What do you want to achieve with deploying functionalities?

[Yongjie]

Thanks for your response. My target device is TI C2000. My target device is the TI C2000, and I am not sure which type of device can realize the computation of CNN Convolutional Networks. Because, usually DSPs have limited computing power. My network is a simple CNN network with two layers. The function that needs to be completed after deployment is to enable online fault diagnosis

[Walter Roberson]

Note that you cannot deploy network training

[Yongjie]

Yes, I know. I just want to deploy the trained model. Training the model can be implemented on the computer.