Tutorial: AI Colorize Black and White with Realism(on Colab)

Introduction

In this tutorial, I'll show you how to transform a black-and-white image into a stunningly realistic color image using a tool called DeOldify. We will go through each step, from setting up the Colab script to tweaking settings for optimal output.

Introduction

What's impressive about the example image we have is the color detail of the Queen of Spades. For curiosity's sake, we’ll take a color image, convert it to black and white, and then use DeOldify to colorize it back. The results are stunning!

Setup and Requirements

Before we start, let’s appreciate the creators of DeOldify. You can find the link to the description of the video as well as the first link that directs you to the Colab script.

Steps:

Copy the Colab Script: Make a copy of the script to your own Google Drive.
Set Session to GPU: Ensure the session is set to GPU for faster processing.

Step-by-Step Process:

1. Setup DeOldify

First, run the initial cell to set up DeOldify on your virtual machine. After the setup completes, there will be instructions to follow at the end of the video.

2. Provide the Black and White Image

You can provide the black-and-white image in two ways:

Using a URL: Find a black-and-white image on the internet, right-click to copy the image address, and paste it into the source URL field in Colab.
Uploading an Image: Upload an image from your device. Click the ‘File’ tab, then the ‘Upload’ button, select your image, and paste its path into the designated field after double-clicking the image.

3. Colorize the Image

Execute the cell after entering the source URL or image path. You will get the colorized version of the image below.

4. Adjust Render Factor

The default render factor is 45, which works well in most cases. However, lower values render faster but with more color inaccuracies. High render factors are better for high-resolution images but may result in slightly washed-out colors. You can experiment with different render factors using the preview cell for all possible values.

Applications and Caveats

This software can be useful for artists, such as creating hyper-realistic portraits and landscapes. Although it’s not perfect, DeOldify does an impressive job at fooling the eyes with vibrant results.

However, there are limitations:

Objects occupying small screen areas may have color inaccuracies.
The tool sometimes misinterprets colors, as seen with a bottle cap being influenced by facial colors.

Understanding the Tech Behind DeOldify

DeOldify initially used Generative Adversarial Networks (GANs), which consist of a generator that creates images from noise and a discriminator that classifies them as real or fake. Later, it switched to self-attention GANs for better continuity and completeness in colorization. Now, it uses non-GAN techniques.

Future Prospects

DeOldify doesn’t just work on images but also on videos. Stay tuned for more updates, and don't forget to like, share, and comment to motivate us to make more high-quality videos.

Thank you!