Turn AI Into a Gallery‑Ready Painter with Docker and Style Transfer

1. A Painter Is Born

In the 2019 graduate exhibition of Central Academy of Fine Arts, a "student" named Xia Yubing, actually an AI robot from MSRA called Microsoft Xiaoice, displayed paintings after over 20 months of training, achieving artistic level comparable to human graduates.

2. Painting Challenges

Before starting, we must set up the environment, installing GPU drivers, CUDA, cuDNN, Torch, protobuf, and loadcaffe for VGG19. Dependency updates can cause crashes, so containerization with Docker and nvidia‑docker isolates the environment while allowing GPU access.

3. Preparing the Painting Tools

We run the project on a GPU‑enabled host, either a private cloud GPU server or a public‑cloud instance such as AWS EC2 P2. The following script installs NVIDIA drivers and nvidia‑docker2.

<code>#!/bin/bash
# Install NVIDIA drivers 418.87
sudo apt-get install --no-install-recommends -y gcc make libc-dev
wget -P /tmp http://cn.download.nvidia.com/tesla/418.87/NVIDIA-Linux-x86_64-418.87.00.run
sudo sh /tmp/NVIDIA-Linux-x86_64-418.87.00.run --silent

# Install nvidia-docker2
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
distribution=$( . /etc/os-release; echo $ID$VERSION_ID )
curl -s -L https://nvidia.github.io/nvidia-docker/${distribution}/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
apt update && apt install nvidia-docker2
sudo pkill -SIGHUP dockerd

# Test
docker run --runtime=nvidia --rm nvidia/cuda:9.0-devel-ubuntu16.04 nvidia-smi</code>

4. Deploy the Model

After building the Docker image, run the style‑transfer script to generate artwork, e.g.,

<code>docker build . -t neural-style:dev-0.0.1
./scripts/fake-it.sh goldengate.jpg monat.jpg</code>

The resulting images can be downloaded or assembled into a GIF.

5. References

A Neural Algorithm of Artistic Style

Visualizing and Understanding Convolutional Networks

How to Fake It As an Artist

neural‑style‑docker

神经美学：发生在大脑和艺术之间的化学反应