Skip to content

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

Notifications You must be signed in to change notification settings

wutianze/dnndk-pynqz2

Repository files navigation

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

About

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

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published