Cartoonizer

Convert images and videos into a cartoon!

The webapp is deployed Zeet.co here - https://cartoonize.zeet.app/cartoonize

The webapp is deployed Heroku here - https://cartoonize-hk.herokuapp.com/

The webapp is deployed Aliyun Severless here - http://cart.ovzv.cn/

---

Update

Use the associated applications of the model

Contents

• Prerequisites for Google Cloud and Algorithmia
• Installation
  • Docker
  • Zeet.co
  • Heroku
  • Aliyun Severless
  • VirtualEnv
  • Google Colab
• API Version
• Sample Image and Video

---

Prerequisites for Google Cloud and Algorithmia

These are important steps if you want to leverage Google buckets, signed URLs and Algorithmia's platform. Skip this if you want to run locally / colab.

Cloud Run authentication

To use any functionalities pertaining to Google Cloud, you'll need a global authentication file (JSON). You can obtain this JSON by following the steps given here - Getting started with authentication

After you get the JSON file, rename it to token.json (so that it's compatible with the codebase).

Set the environment variable in your terminal -

export GOOGLE_APPLICATION_CREDENTIALS="/path/to/token.json"

Notes:

• You can set it permanently by adding this line to ~/.bashrc.
• Dockerfile already includes the setting of this particular environment variable. :)

Algorithmia

We used the Serveless AI Layer product of Algorithmia for inference on videos. To learn more on how to deploy your model in Algorithmia, check here - https://algorithmia.com/developers

---

Installation

Application tested on:

• python 3.7
• tensorflow 2.1.0
• tf_slim 1.1.0
• ffmpeg 3.4.8
• Cuda version 10.1
• OS: Linux (Ubuntu 18.04)

Using Docker

The easiest way to get the webapp running is by using the Dockerfile:

1. cd into the root directory and build the image

docker build -t cartoonize .

Note: Set the appropriate values in config.yaml before building the image.

2. Run the container by exposing the appropriate ports

docker run -p 8080:8080 cartoonize

Using Zeet.co

• https://zeet.co/home
• CPU and GPU Build Success https://hub.docker.com/u/literature

1. Before starting, you need to log in to zeet and log in directly with your Github account.

2. Choose docker image

3. Choose a free plan.The free plan can only deploy one CPU container, the second one requires payment, the free plan only has CPU, and the paid plan has GPU.

4. You can go to the address(https://hub.docker.com/r/literature/cartoonize-cpu/tags), of course, you can also compile by yourself.

• The difference between CPU and CPU_15s is that CPU_15s can only process videos up to 15s, while cpu does not limit the length of the video.

5. Fill in the image name you want, Zeet.co will automatically pull it and complete the deployment.

• such as literature/cartoonize-cpu:cpu

6. You can get the deployment process in real time, and click Visit Active Deployment after completion Visit your application.

7. Zeet provides a free container for each user, please don’t abuse it and cherish resources.

Using Heroku

• https://dashboard.heroku.com/apps
• There are some differences between the general docker image and the docker image used by Heroku, but I have compiled the image used by Heroku and can use it directly.
• CPU and GPU for Heroku Build Success https://hub.docker.com/u/literature

1. Create a new app, choose any name, but don’t use the same name as others

2. After entering the app, select deploy, and then select Container Registry, Pay attention to the following command, and learn more on the left, there are many commands you will use, if there is an error, you can check the command yourself

3. I use my own compiled image to deploy, you can use it, or you can compile one yourself.

• I use my own server for deployment. If you want to deploy a server, you need to make sure that your local network is consistent with the server network so that Heroku can verify your identity. There is no such problem when using local.
• Make sure your server environment or local environment has Docker and Heroku CLI installed, If not, please install it first. These two conditions determine whether it is successful. Give two links for reference:
• Docker
• Heroku CLI

4. Pull the mirror to the local

docker pull literature/cartoonize-cpu:heroku

5. Log in to Heroku.If you use the server deployment, you will be given a link, click the link to verify, but the server network and your local network must be the same, if you use the local PC deployment, it will automatically jump, just click verify.

heroku login

6. Logging in to the registry. Heroku runs a container registry on registry.heroku.com.

heroku container:login

7. Pushing an existing image to your Heroku app.

• Official order

docker tag <image> registry.heroku.com/<app>/<process-type>
docker push registry.heroku.com/<app>/<process-type>

• Modify it like me, is changed to literature/cartoonize-cpu:heroku, and app is changed to the name of the application we created in the first step. <process-type> is modified to the type you want to publish. I will modify it to web here. For more information, please refer to the official documentation.

docker tag literature/cartoonize-cpu:heroku registry.heroku.com/cartoonize-hk/web
docker push registry.heroku.com/cartoonize-hk/web

8. Now we push the image to the Heroku registry, and then load the warehouse image into the application.

• The parameter behind -a is your application name

heroku container:release web -a cartoonize-hk

9. Launch your app.

• The parameter behind -a is your application name

heroku open -a cartoonize-hk

• Copy the link to the browser to access

10. Heroku also provides free services to everyone, please don't abuse it.

Using Aliyun Severless

1. Register for Aliyun

2. Container image service

3. Create a namespace

4. Create image warehouse

5. Name the compiled image as required. Push to warehouse after naming

• Log in to the warehouse (these commands are available in alicloud)

• Rename the image to the format required by alicloud

• You can compile by yourself, and I also provide the version for Aliyun Severless on docker hub docker pull literature/cartoonize-cpu:sf.

• Push image

6. Open the serverless computing console

7. Create an app

8. Fill in the relevant configuration. The image was just pushed by us

9. Just a moment to deploy, Access test domain name

10. Bind custom domain name . The path must be filled in /*

11. Provides many functions, such as memory expansion, concurrency settings, etc., can be modified

Using virtualenv

1. Make a virtual environment using virutalenv and activate it

virtualenv -p python3 cartoonize
source cartoonize/bin/activate

2. Install python dependencies

pip install -r requirements.txt

3. Run the webapp. Be sure to set the appropriate values in config.yaml file before running the application.

python app.py

Using Google Colab

1. Clone the repository using either of the below mentioned way:

   • Using Command:

     • Create a new Notebook in Colab and in the cell execute the below command.

      ! git clone https://github.com/experience-ml/cartoonize.git
     

     Note: Don't forget to add ! at the beginning of the command

   • From Colab User Interface

       Open Colab
           └── File
                └── Open Notebook
                         └── Github
                               └── paste the Url of the repository

Note : Before running the application change the runtime to GPU for processing videos but you for images CPU shall also work just fine.

           Runtime
              └── Change runtime type
                          └── Select GPU

2. After cloning the repository navigate to the /cartoonize using below command in the notebook cell:

   %cd cartoonize
   

3. Run the below commands in the notebook cell to install the requirements.

   !pip install -r requirements.txt
   

4. In config.yaml file set:

   colab-mode: true 
   

5. Launch the flask app on ngrok

   !python app.py
   

Note : Sample Google Colab Notebook for reference

---

API Version

• Use the following docker command to deploy the server

docker pull literature/cartoonize-cpu:api
docker run -p 8080:8080 literature/cartoonize-cpu:api

or

docker pull literature/cartoonize-gpu:api
docker run -p 8080:8080 literature/cartoonize-gpu:api

• The service will run on your local port 8080 or other ports you specify. If you need domain name access, you can act as a proxy.
• provided that you have installed docker. After you deploy correctly, both GET and POST requests can be accessed. The actual display is as follows
  • Interface: http://your domain/api or http://127.0.0.1:8080/api can be accessed.
  • Parameter: image value: a picture
  • Return value: the base64 data stream after processing the image

Sample Image and Video

Emma Watson Cartoonized

Youtube Video of Avenger's Bar Scene Cartoonized

---

License

1. Copyright © Literature (Demo webapp)

   • Authors: Literature.

   • Licensed under the CC BY-NC-SA 4.0

   • Commercial application is prohibited by license

2. Copyright (C) Xinrui Wang, Jinze Yu. (White box cartoonization)

   • All rights reserved.
   • Licensed under the CC BY-NC-SA 4.0
   • Also, Commercial application is prohibited license (https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode).