Seam carving is a technique that can be used to resize images, which is also known as liquid rescaling. Compared to the traditional resizing tool, it does not “stretch” the image, but it selectively removes the pixels which contain the least amount of information. As a result, it allows to shrink images preserving most of the details.
Seam carving is an example of a context-aware resizing algorithm, as it does not treat images as mere collections of pixels. By all means, it can be considered an AI-powered algorithm. The “AI part” resides in the fact that it is able to identify which pixels to remove on its own. However, it does so without any neural network and—most importantly—without the need to be trained on external data. Hence, it belongs to the field of what I call Classical AI, conversely to the more recent field of Deep Learning. With AI-powered tools becoming more and more popular, I find it helpful to show how a lot can be achieved with some clever algorithms, without the need to train expensive neural network models.
If you are interested in learning more about tools like DALL·E 2 and Midjourney, I would suggest checking one of my most detailed articles titled The Rise of AI Art.
This post continues our journey through the Mathematical foundations of iridescence. This time, we will discuss a new way in which material can split light: thin-film interference. This is how bubbles (and car paint) get their unique reflections.
In this fifth post, we will recreate the shimmering reflections that are typically seen on sand dunes.
Shortly after the publication of this series, Julian Oberbeck and Paul Nadalack made their own attempt at recreating a Journey-inspired scene in Unity. You can see in the thread below how they have improved the glitter reflection to have more temporal coherence. You can read more about their implementation on IndieBurg’s article Mip Map Folding.