MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed Meschers tool for a novel geometry representation that enables processing and analysis of physically impossible objects—optical illusions that defy conventional 3D logic, such as Penrose triangles and Escher-style constructions, informs MIT News.
The key innovation is its use of a 2.5D mesh model, which stores 2D image-plane positions (x, y) augmented with relative per-edge depth shifts (Δz) instead of full 3D coordinates. This preserves local visual plausibility without enforcing global geometric consistency, mirroring how the human visual system perceives impossible shapes.
Geometry-Processing Features
- Subdivision and smoothing apply classic mesh refinement techniques to reduce blockiness and improve the appearance of impossible surfaces while maintaining the illusion.
- Geodesic computations allow calculation of shortest paths across the surface of an impossible object (e.g., an “impossibagel”)—enabling distance queries as if the shape were real.
- Physical simulations, including heat diffusion, function over the mesh topology—even on surfaces that cannot exist in conventional 3D.
- Inverse rendering tools convert 2D drawings or target images into editable meschers, facilitating transformation from flat images into manipulable, illusion-retaining meshes.
Under the hood, Meschers’ representation is grounded in discrete exterior calculus, permitting intrinsic operators (like the Laplacian) to work despite the object’s global incoherence. This enables robust geometry processing aligned with perceptual cues rather than physical constraints.
With Meschers, MIT researchers offer a bridge between perceptual illusion and computational geometry, empowering designers and scientists to apply real geometry algorithms to shapes that will never exist physically.
Lead author and MIT PhD student Ana Dodik says, “Using Meschers, we’ve unlocked a new class of shapes for artists to work with on the computer. They could also help perception scientists understand the point at which an object truly becomes impossible.”
Dodik and her colleagues will present their paper at the SIGGRAPH 2025 conference.