Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
7 years ago

Object detection on DE1/DE10-Nano FPGA SOC

I have re-trained SSD-Mobilenet using Tensorflow framework on my own image dataset and able to use it on Ubuntu/Windows with CPU,GPU. I want to test performance on the FPGA board either DE1/DE10-Nano FPGA SOC. Is it possible to use my pre-trained tensorflow model on these boards ? If possible could you guide the workflow i need to take up to achieve this?

2 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    7 years ago

    Well, the only available open-source OpenCL based FPGA design for CNN inference is pipecnn (https://github.com/doonny/pipecnn)

    and it only support AlexNet and VGG for now.

    For SSD and MobileNet, you need dilate convolution and separable convolution, neither is supported by PipeCNN.

    Either wait for Intel to release CNN library for FPGA, or implement those missing parts based on PipeCNN (actually I'm also trying to do that).