SCI-FI PALETTE

🚧 Work In Progress 🚧

 

Overview

I explored a movie poster database with image clustering and dimension reduction (UMap technique) in machine learning. And I made an interactive website out of it.

Due to the time limitation and performance of the web, I used only 2,000 poster samples in my most interested genre --- science fiction as the input dataset. Although not flawless, some salient patterns in sci-fi movie posters are revealed --- colors, elements, compositions, layouts, visual styles and release years...

 
 

:/ Machine Learning & Experimental Web

:/ Python, HTML+CSS+Javascript, PixiJS

:/ Individual work

:/ 2020.03- One Week

 

Click the button below to the project website or go to http://movieposterml.waynecai.com/

Movie poster data source: www.movieposterdb.com

This web application could be used as a practical tool for designers to learn poster design through movies, or just for a movie fan to roam in a movie poster galaxy!

 
 

KEY PROCESS

 

Main implement process of this project includes data scraping and cleaning, image resizing, extracting color palette from images, and using pre-trained neural network for image clustering.

 

KMeans function & 'Elbow method'

Dimensionality Reduction -UMAP (6000 data points)

Convert the point cloud into a 2d grid using Rasterfairy.(6000 data points)

FUTURE STEPS

 

  • Including all data points from the sci-fi movie database;

  • Identifying elements in posters with object detection, to discover movie poster in detail.

Inspirations: Beyond Scrolls