README.md

Video Color Spectrum

This repository contains the script to extract the color spectrum of a YouTube video using defined algorithm configs. The pipeline for a single video can be divided into 4 steps:

1. Download an YouTube video
2. Break the video into its frames
3. Apply smoothing pre-processing to the frames
4. Extract the most predominant colors for each frame

An extensive explanation of the pipeline can be found in this medium article.

Example

By applying the pipeline for this video, we get the following color spectrum:

Running the pipeline

Install dependencies

The pipeline uses libraries specific libraries like:

• PyYAML: reads configuration files
• opencv-python: video and image manipulation
• mlflow: experiments tracking

To install all dependencies, run:

pip install -r requirements

Execute pipeline

To run the pre-defined configs, execute the following command:

make run

Results are stored on local files but they are also saved using the experiments structure of the mlflow framework. To start an mlflow server and see the results, run the following command:

make server

Define alternative configs

Once you successfully run the pipeline for the standard configs, you can apply it to other YouTube videos with other pipeline parameters. To do so, edit the configs.yaml file which is structured as follows:

# Smooth preprocessing parameters
smooth:
  method: GaussianBlur # options: AverageBlur, MedianBlur, GaussianBlur
  kernel_size: 1

# Clustering parameters
color_extraction:
  method: KMeans # options: KMeans, DBSCAN
  params:
    # Sklearn clustering parameters
    n_clusters: 3

-
    output_folder: abertura-hora-1 # extraction experiment/output folder
    url: https://www.youtube.com/watch?v=BNaT9M5-Q1o # YouTube URL video
    resolution: 144p # video resolution (defaults to 144p)