Skip to content

Latest commit

 

History

History

interactive_demo

README.md

GAN Compression Interactive Demo on Jetson Nano

Overview

We show how to deploy our compressed Pix2pix model on NVIDIA Jetson Nano. Our model is compiled with TVM Auto Scheduler [1,2] for acceleration. The final model achieves about 8 FPS on Jetson Nano GPU.

[1] Tianqi Chen et al., TVM: An automated end-to-end optimizing compiler for deep learning, in OSDI 2018

[2] Lianmin Zheng et al., Ansor: Generating High-Performance Tensor Programs for Deep Learning., in OSDI 2020

Getting Started

  1. Get an NVIDIA Jeston Nano board (it is only $99!).

  2. Get a micro SD card and burn the Nano system image into it following here. Insert the card and boot the Nano. Note: you may want to get a power adaptor for a stable power supply.

  3. Follow here to install PyTorch and torchvision.

  4. Install TVM 0.8.

    # upgrade cmake
cd ~
sudo apt install openssl libssl-dev
sudo apt remove cmake
wget https://github.com/Kitware/CMake/releases/download/v3.21.0-rc2/cmake-3.21.0-rc2-linux-aarch64.sh
sudo bash cmake-3.21.0-rc2-linux-aarch64.sh --prefix=/usr --exclude-subdir --skip-license

cd ~
sudo apt install llvm # install llvm which is required by tvm
git clone --recursive https://github.com/apache/tvm tvm
cd tvm
mkdir build
cp cmake/config.cmake build/
cd build
vim config.cmake
# edit config.cmake to change
# USE_CUDA OFF -> USE_CUDA ON
# USE_LLVM OFF -> USE_LLVM ON
# USE_THRUST OFF -> USE_THRUST ON
# USE_GRAPH_EXECUTOR_CUDA_GRAPH OFF -> USE_GRAPH_EXECUTOR_CUDA_GRAPH ON

cmake ..
make -j4
cd ..
cd python; python3 setup.py install --user; cd ..

  5. Install PyQt5 (Jetson Nano may have pre-installed it).

  6. You now can try our demo with

    python3 paint.py

Acknowledgement

This demo code is developed based on Piecasso.