Pixel Wrapper

A wrapper to run Pixel.js on top of Diva.js as a job in the workflow builder Rodan

Installation

• Follow the rodan-docker guide to have docker set up.

• Clone this repo inside the rodan jobs folder (rodan-docker/jobs) using

  git clone --recurse-submodules https://github.com/DDMAL/pixel_wrapper.git

  • If using an older version of git (pre-2.13) and the above command fails, instead run

    git clone https://github.com/DDMAL/pixel_wrapper.git
    git submodule update --init --recursive

  • If you already have an outdated version of this repository cloned, then pull all the changes and run

    git submodule update --init --recursive

• Copy jobs/setting.py.job_development to jobs/settings.py.
• Open up jobs/settings.py in a text editor. You'll need to do two things:
  • Include the path to the wrapper folder in the Rodan Job Package registration (line 96). This should look something like the following

    RODAN_JOB_PACKAGES = (
    "rodan.jobs.pixel_wrapper",
    # Paths to other jobs
    )

  • Increase RODAN_RUNJOB_WORKING_USER_EXPIRY_SECONDS (line 87) from 15 to 150000.
• In RODAN_JOB_PACKAGES, check if rodan.jobs.pil-rodan is included in the job paths.
  • If pil-rodan is not in the list, clone https://github.com/DDMAL/pil-rodan.git to the jobs folder, like in the first step and add its path to the list of rodan job packages like so:

    RODAN_JOB_PACKAGES = (
    "rodan.jobs.pixel_wrapper", 
    "rodan.jobs.pil-rodan",
    # Paths to other jobs
    )

• Open docker-compose.job-dev.yml and replace both occurences of demojob with pixel_wrapper.
• Setup Pixel:
  • At the root of the project, run python activate_wrapper.py in terminal to activate the wrapper within Pixel.js.
  • In source/js/plugins/pixel.js run ./pixel.sh in terminal to install all dependencies and compile the project.
• The wrapper should now be available to use in any workflow

Running Rodan

• Once the above installation steps are complete, run Rodan with the following command:

  docker-compose -f docker-compose.yml -f docker-compose.job-dev.yml up

Using Pixel as a Rodan job

Here are some user-level instructions on adding a pixel.js job in RODAN.

• By following the rodan-docker guide, the rodan client should now be running on localhost:9002
• Create a new workflow, or select an existing one.
• Double-click workflow -> workflow header tab -> add job
• Find Diva - Pixel.js -> Add
• Double-click red input square to select your image from resources (you must pre-upload resources, select the resource section on the left to do so)
• Choose png or tiff resource -> add -> input square should be green now
• Workflow header tab -> run

Port setup

• You must create however many layers that you wish to classify as the input ports.
  • Do this by right clicking on the Pixel job, clicking on ports, and adding/deleting input ports as needed.
  • For example, if you're working with 3 layers (music symbols, text, and staff lines), then create three input ports.
• There must be two more output ports than input ports.
  • This is because one port is reserved for the auto-generated Background Layer (the negative of the classified layers combined), and another is reserved for the Select Regions Layer (a mask of the user-verified portions of the image).
  • Ensure that your output ports are in incremental order (ie. do not skip from layer 1 to layer 5).
• For example, if you have Layers {1, 2, 3} as input ports, then you must have Layers {0, 1, 2, 3, 4} as output ports. Layer 0 is already enabled by default, so you'd just need to enable the other 4.

Making changes to the pixel_wrapper source code

Sometimes changes need to be done to the wrapper code found in source/js/plugins/pixel-wrapper.js, or the Pixel.js source code in source/js/plugins/Pixel.js.

If this is the case, make sure to run gulp develop:rodan from the pixel_wrapper directory after making any changes. This will compile the source code and move it to the static folder, which is used to upload the code to the server running Rodan. You won't need to restart Rodan.

If you make any changes to the css files, you'll need to move them manually the pixel_wrapper/static/css folder.

Testing

Unit testing for Pixel Wrapper is done using Selenium, and requires a web browser (FireFox, make sure you have the latest version if already installed), and its driver (geckodriver).

• Once you install geckodriver, place it in your system PATH, perhaps like so

  mv ~/Downloads/geckodriver /usr/local/bin

• You can access the pixel-wrapper.test.js file in source/js/plugins in order to add/remove/alter existing tests.
• Once ready to run the tests, start the server with gulp. It must be running for the unit tests to work. You should either run it as a background process (gulp &), or in another terminal.
• Run npm test.

Differences from the standalone Pixel

This wrapper changes some of the functionality inherited from the standalone Pixel.js, as well as introducing some new ones.

• Creating and deleting layers
  • This functionality was removed in order remove any possible user-generated conflicts with the number of output ports and the number of actual layers.
  • The number of layers will be determined by the number of input ports created for the Pixel.js job; if none, then the user will be prompted and must accurately determine the number of layers to classify.
• Select region layer
  • This is the default layer selected when opening a Pixel.js job. The user should use this layer to select the regions of the page that they will classify, and should stick to within these regions.
• Export buttons
  • The [Submit To Rodan] button allows the user to submit the classified layers to the Rodan workflow.
  • The [Export as CSV] and [Export as image Data PNG] buttons, while functional, will cause the exported layers to have streaks of empty image data due to the filetype change required for compatibility with the Convolutional Method. These buttons in general shouldn't be used, as they are not required within any workflow.
• Background layer
  • This layer is automatically generated when using the [Submit To Rodan] button. It will be generated as the negative (or difference) of the other layers within the selection regions.