Marginaliadaily

Creating 3D Optical Illusions with AI

orig. “JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-Space Denoising” · Siang-Ling Zhang, Huai-Hsun Cheng, Tsung-Ju Yang, Yu-Lun Liu

Computer Vision Intermediate 5 min read AI-assisted, reviewed by Alex Dong
In the margin
Area
Computer Vision, Getting computers to understand images and video — recognising objects, scenes, and actions.

Imagine being able to create stunning 3D visual illusions that change as you look at them from different angles, and now AI can do it quickly and easily

3D visual illusions are like magic tricks for your eyes, where one object looks like two different things depending on the angle you view it from.

Creating these illusions used to be a slow and difficult process, with existing methods producing unrealistic colors or visible seams.

The new approach, called JanusMesh, uses a two-stage process to generate these illusions. First, it creates a 3D shape that can be viewed from different angles, and then it adds textures to the shape to make it look realistic from each angle.

This process uses denoising, which is like cleaning up noise in an image, and diffusion priors, which help the AI understand what the texture should look like from different views

The ability to create realistic 3D visual illusions quickly and easily has many potential applications, from art and design to education and training. It could also be used to create more realistic special effects in movies and video games, or to help architects and engineers visualize and communicate their designs more effectively

Source

Siang-Ling Zhang, Huai-Hsun Cheng, Tsung-Ju Yang, Yu-Lun Liu

We write original plain-language summaries and link to the source. We never republish the paper.

Paste it and we'll explain it even more simply.

Pass all three to earn the “read & understood” stamp (+10 pts).

Member notes Sign in ↗
AD Alex Dong TEAM
We read the full paper and rewrote it in plain language. Leave your own note below.
Leaderboard · this week

Pass quizzes and leave notes to climb your chapter's board. No chapters are running yet, so this one is wide open.

Start a chapter to compete →
A Smarter Way for AI to Understand Videos
Intermediate
Making Sense of Long Videos
Intermediate
Teaching AI to Understand Videos from a First-Person View
Intermediate