STUDY 0: Study on possible palette to story correlations in anime trailers

Assumptions

• There is a correlation between color palette and genre of anime

Gathering of Dataset

Trailer

Collect trailers of at least 1000 different anime. Several seasons of the same major anime story line are considered different anime.

Meta Information

• Title
• Feature
  • Release date
  • Length of trailer in frames
  • Genre tags
  • AniDB tags
  • AniList tags
  • Rating

Scope

• Color anime, no black and white renderings
• No anime older than 2000

Random Split of Dataset

The dataset will be randomly split into a teaching set and a verification set. The teaching set will be used to teach one or more classificators which will have to be defined. The verification set will be used to verify or falsify assumptions made based on the previously tought features.

Processing

Meta Information

Tags describing the same feature characteristic but differing in tag representation (eg. "Mecha", "mecha", "Mecha-Anime") will be consolidated. Tags describing more than one feature characteristic will be split up (eg. "Space/Mecha" will be "Space" and "Mecha").

Data processing

Each frame of a given trailer will be dumped to an individual image. Images will then be analyzed based on color histograms using a fixed bin width across the whole study to determine by-frame color distribution. The study will focus on the HSV color space. We'll follow the basic methods outlined in Oranges and Blues and adapt them to anime.

Installation Instructions

This study needs python3 and virtualenvwrapper (sudo pip install virtualenvwrapper). Also current implementation requires a PostgreSQL DB (credentials must be changed in xmldump_to_db.py)

• Create a virtualenv:

  mkvirtualenv --no-site-packages --python=$(which python3) studyzero

• Activate this environment:

  workon studyzero

• Then cd to where the code lies and run

  sudo apt-get install python3-dev libpq-dev  # so pip can install psycopg
  pip install -r requirements.txt

  • You can also setvirtualenvproject here to set the current directory as project directory for the virtualenv. 'workon studyzero' will then automatically change into this directory.
• Download 'anime-titles.xml.gz' from anidb. (I am not linking to this file so they won't get hammered needlessly)
• Change 'xmldump_to_db.py' to fit your credentials.
• Run

  python xmldump_to_db.py

  and wait, for this might take a few seconds

• You now have about 8000 anime with their associated titles in your PostgreSQL DB!