Light Field Lab

What this model represents

• Real: rays come from the cube. Flat: a fixed perspective picture is on a panel 0.65 m from the initial eye position. Light field: an ideal directional surface reproduces rays from a virtual cube behind that panel. The dashed portions are not real light behind the display.
• The light field surface represents controlled light position and direction, not an ordinary emissive panel. It does not model specific pinholes, microlenses, elemental pixels, finite angular sampling, diffraction, or crosstalk. Drawn ray count changes the illustration, not device resolution.
• Pinhole array: a separate geometric construction. For each marked cube point and aperture, an associated rear pixel is found by extending the required straight ray backward. The mask selects directions; it does not bend rays. Only rays crossing the enlarged pupil enter the schematic eye. Rear pixels represent samples of elemental images, not one intact cube picture.
• Pinhole mode always includes finite-aperture blur. The ray slider controls samples across each optical opening. Aperture count defaults to 120. Optical aperture width is fixed at 0.20 mm with a 22 mm display-to-mask gap. This is a representative submillimeter aperture, not an optimized hardware specification: selecting an optimum requires wavelength, pixel size, and diffraction modeling. Width uses the scene's meter scale and stays fixed when aperture count changes; the visibly enlarged mask gaps do not determine optical width.
• Rear pixels are restricted to their associated elemental-image cells. Ray counts are illustrative, not headset specifications: the eye and pupil are enlarged. Finite width illustrates geometric angular spread only. Diffraction, radiometric brightness loss, and leakage through neighboring apertures are not simulated.
• Display-to-pupil: uses the same enlarged-eye geometry and pixel-to-aperture paths as Object rays. A teal fan bounds paths from the selected elemental-image cell through its aperture center; it is light coverage, not an image projected into space or a guaranteed full-image eyebox. Neighboring fans are only context. Solid red/blue paths reach the pupil; dashed paths miss it. The lower inset magnifies one cell uniformly and shows top-view projections of the cube vertices onto the rear display, not a full two-dimensional elemental image. Focus changes the eye image, not these incoming paths. Finite-width ray samples are included, but the coverage fan remains a center-ray envelope. No extra mask or dual-scene hardware is introduced.
• The pinhole mode's eye view shows the ideal reconstructed target, using the same finite-pupil rendering as ideal light-field mode. It does not predict the finite array's artifacts, brightness, or sparse-ray accommodation behavior. Most light from a real pixel is blocked by the mask. Smaller holes can improve angular selectivity but reduce throughput; diffraction eventually matters.
• The eye is a rotating spherical cross-section. Its size and pupil are exaggerated. A thin-lens calculation gives the image-plane intercept; this is projected from the lens center onto the drawn curved retina. This idealized curved-image mapping avoids artificial field-curvature blur caused by the enlarged diagram. Only the selected focus plane is sharp; anatomical aberrations are not simulated.
• The upright eye-view estimate uses Three.js and 32 finite-aperture samples focused at one distance. Defocus is exaggerated 2x for classroom visibility by doubling the sampled aperture in this preview only. The selected focus distance, nominal pupil setting, and center ray diagram are unchanged. It shows a cropped 55-degree horizontal field, not the eye's entire field of view. It is an illustrative optical image, not a prediction of subjective perception. A flat picture can still look like a cube through pictorial depth cues; its optical focus remains on the panel.
• Gaze and accommodation are independent controls. A/B gaze presets align the horizontal direction only. One eye is modeled: binocular disparity, vergence, and neural processing are not simulated. Eye shift means motion relative to a stationary display, not tracked head movement with a worn headset.

References: Lanman & Luebke, Near-Eye Light Field Displays (2013); Viewing-Angle-Enhanced and Dual-View Compatible Integral Imaging 3D Display Based on a Dual Pinhole Array (2024). Real systems have spatial/angular resolution and depth-of-field trade-offs not reproduced by the ideal mode.

Three.js r160, MIT license (bundled for offline use). Original simulator supplied by the instructor; revised geometry, eye model, and rendering.