U-NET implementation on PYNQ-Z2 for Lung CT Segmentation

Support
1

I am working on a project that uses the U-Net deep learning model for lung CT image segmentation (lung tumor detection).

My goal is to deploy or accelerate the inference on the PYNQ-Z2 FPGA board using the ARM + FPGA architecture.

Currently, I would like to understand:

  • Is it feasible to implement U-Net on PYNQ-Z2?

  • What is the recommended workflow for deploying a PyTorch model on PYNQ?

  • Should I use Vitis AI, HLS, or another framework?

  • Are there any examples of CNN/U-Net acceleration on PYNQ-Z2?

Project details:

  • Model: U-Net

  • Dataset: Lung CT images

  • Framework: PyTorch

  • Target board: PYNQ-Z2

  • Task: Image segmentation

Any advice, tutorials, or example projects would be greatly appreciated.

Thank you.

2

Hi @BENMADANI_ABDERRAHMA, welcome to the PYNQ community.

I would suggest looking at the FINN framework for implementing ML models on FPGAs. Additionally, there are actually quite a few research papers out there doing ML image processing using PYNQ. Have a look for relevant keywords on IEEE Xplore for inspiration.

3

Can a Tiny U-Net architecture be integrated into the FINN framework for deployment on PYNQ-Z2?

4

Hi @BENMADANI_ABDERRAHMA,

You may be better off asking questions about FINN in their GitHub Discussions channel. Xilinx/finn · Discussions · GitHub