pramod627 / wutianze_dnndk-pynqz2

First lesson for you to use DNNDK, also it can be helpful for your AI learning

Edge AI Tutorials

Zynq 7000 DPU TRD

dnndk3.0-pynqz2

• In this tutorial you will learn:

  1. How to use caffe model resnet50 to classify pictures using pynq-z2.
  2. How to use tensorflow model mnist to recognize hand-writing number using pynq-z2.
  3. How to train and use yolov3 in pynq-z2.
  4. How to use DNNDK-v3.0 to optimize the trained models.
  5. How to use dpu in pynq-z2 to accelerate inference.

• First download all the files to your pc.

  You can also download the system image of pynq-z2 we provided in Baidu Cloud or Google Cloud(If some files are missing, pls find them here), it embeds DPU IP into pynq system and fixes some problems of official image. For more details, please refer to HydraMini.

• The most important files are organized as followed:

  mnist_tf

  mnist_host
  mnist_pynqz2
  mnist-handwriting-guide.md

  resnet50_caffe

  resnet50_host
  resnet50_pynqz2
  resnet50_pynqz2_guide.md

  yolo_keras

  keras-yolo3
  yolo_pynqz2
  take_training_imgs
  yolo_pynqz2_guide.md

  The mnist_tf contains the mnist model trained by tensorflow and you can read the mnist-handwriting-guide.md to learn. The resnet50_caffe contains the resnet50 model trained by caffe and you can read the resnet50_pynqz2_guide.md to learn. The yolo_keras provide a yolo implementation using keras, you can download the pre-trained weights of yolo from darknet.

• Preparation

  Before you start, you should read build-host-dnndk.md & build-pynqz2-system.md first to set your environment and do some preparation. I recommend you learn mnist_tf before running into yolo_keras.

If you have any problem, please open an issue. If you like this project please star to support.

Here is a wonderful practice of this project presented by André Dias Araújo, it implements a YOLOv3 object detector on a PYNQ-Z2 PLD, making use of the Zynq-7020 to accelerate the inference and provide accurate results.

You wanna something more exciting? See HydraMini