Vis2Touch: Perceptually Grounded 3D Printing of Visual Art With AI-Driven Depth and Texture Estimation

Visual art offers a rich expression of its subject's color and form, but lacks the haptic cues of depth and texture central to understanding its physical characteristics. We present Vis2Touch, a system that transforms color images into textured 2.5D models for 3D printing by combining depth and surface texture while preserving the original color appearance. Vis2Touch composes 2.5D objects by combining a monocular depth estimator to reconstruct macroscale depth and a fine-tuned diffusion model to generate microscale textures from semantically segmented image regions. These two heightfields are superimposed to produce composite colored 3D models that reflect both the structure and feel of the source image. Through psychophysical experiments informed by a gallery exhibition, we investigate how tactile expectations vary with depth and show that perceptual accuracy improves when texture is modulated as a power-law function of distance in the image. Our results demonstrate new possibilities for physicalizing visual art by conveying spatial scale through 3D-printed haptic artifacts.