Training a very light model on Up board

dhoa
dhoa New Member Posts: 13
edited February 2019 in UP AI Core Videos

Hi everyone. I'm a new one here and might be I put the question in a wrong place. If yes please tell me where I should go, thank you.

I have some Up boards and working on a computer vision problem. I know that Up board is made for inference only and not for training. But can we just train a very very light deep learning model ? If yes which framework is supported (tensorflow, pytorch, ...) and how I can install it ? How about other machine learning model (random forest, ...)

Thank you for your help,
Regards,

Tagged:

Comments

  • Ethan_Xu
    Ethan_Xu New Member Posts: 1

    Do you receive any answers? I'm also very curious about that.

  • DCleri
    DCleri Administrator, AAEON Posts: 1,213 admin

    You can try to run a training with tensorflow using the CPU, but you would need a build without AVX instructions. Also the performance would be extremely low.

    On UP Squared and newer you can use tensorflow-opencl and run training using the integrated GPU, also this one was not tested, but it should be working using the correct tensorflow package.