July 16, 2026
NeLU3D Paper Accepted to ECCV 2026
News2026-07-16
The paper 'NeLU3D: Neural Inverse Structured Light without Modeling the Projector' has been accepted to ECCV 2026.
The paper "NeLU3D: Neural Inverse Structured Light without Modeling the Projector," authored by Giancarlo Pereira, David Fouhey and Daniele Panozzo, has been accepted to ECCV 2026.
The paper rethinks structured light triangulation, a workhorse technique for precise 3D shape acquisition that traditionally requires careful geometric and radiometric calibration of the camera-projector pair. NeLU3D is a neural inverse structured light method that skips explicit projector modeling entirely: with as few as four monochromatic images (or two RGB images), it recovers accurate surfaces and normals using arbitrary light sources, from extremely low-cost projectors to high-speed analog projectors for slow-motion capture. Across more than twenty-five objects of varying shapes, reflectances, and textures, it achieves sub-millimeter accuracy even with suboptimal patterns where previous methods fail or produce noisy outliers.
The paper rethinks structured light triangulation, a workhorse technique for precise 3D shape acquisition that traditionally requires careful geometric and radiometric calibration of the camera-projector pair. NeLU3D is a neural inverse structured light method that skips explicit projector modeling entirely: with as few as four monochromatic images (or two RGB images), it recovers accurate surfaces and normals using arbitrary light sources, from extremely low-cost projectors to high-speed analog projectors for slow-motion capture. Across more than twenty-five objects of varying shapes, reflectances, and textures, it achieves sub-millimeter accuracy even with suboptimal patterns where previous methods fail or produce noisy outliers.