This Algorithm Can Reconstruct 2D Sketches in 3D

This symmetry-driven is capable of recovering designer-intended 3D geometry from concept sketches.

During SIGGRAPH 2022, an independent team of researchers demonstrated Symmetry Sketch, a useful symmetry-driven algorithm capable of turning 2D sketches into 3D versions of themselves. According to the team, the method was designed for recovering designer-intended 3D geometry from concept sketches, as such sketches, while being inherently imprecise, are highly effective at communicating 3D shapes to human observers.

The team's presentation states that the algorithm first decomposes the sketch into locally-symmetric groups of strokes and proceeds to identify pairs of strokes that are symmetric with respect to triplets of axis-aligned planes. At its core, the method selects the stroke correspondences that result in the most symmetric and well-connected shape.

And here are some examples:

"We observe that most concept sketches of human-made shapes are structured around locally symmetric building blocks, defined by triplets of orthogonal symmetry planes. We identify potential building blocks using a combination of 2D symmetries and drawing order. We reconstruct each such building block by leveraging a combination of perceptual cues and observations about designer drawing choices. We cast this reconstruction as an integer programming problem where we seek to identify, among the large set of candidate symmetry correspondences formed by approximate pen strokes, the subset that results in the most symmetric and well-connected shape," stated the team.

You can learn more about the algorithm here and access its code over here. Also, don't forget to join our Reddit page and our Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more. 

Join discussion

Comments 0

    You might also like

    We need your consent

    We use cookies on this website to make your browsing experience better. By using the site you agree to our use of cookies.Learn more