A wrong rendered IPD globally changes size judgments

What it is

The distance between the two render cameras is rendered / virtual IPD, not the physical spacing of the headset lenses. Cameras wider than the wearer’s true IPD make the same disparity read as “nearer and smaller” — the room collapses to a model, Lilliputism. Cameras narrower, and the world inflates. One wrong setting shifts size globally; it is not a few objects with the wrong texture scale.

Lenses off the pupils blur and fatigue; that is a different path. Here the image can be sharp and the measuring stick still slides, because the geometric baseline is wrong.

Why it happens

Stereopsis inverts depth assuming the baseline is one’s own IPD. Retinal disparity δ is roughly IPD / D, and D is recovered from that. Render with IPD′, recover with IPD, and the perceived distance is true distance times IPD/IPD′. Size tracks distance: a table of fixed angular size, recovered nearer, becomes a toy; recovered farther, a giant prop.

This is not “scale the mesh down”. Mesh scale changes objects relative to the world; rendered IPD changes the observer’s ruler. Floor texture, furniture and the virtual hands move together, so people cannot recalibrate from “this cup looks wrong” — the whole stick is bent.

A runtime that hands everyone a population mean (say 63 mm) parks people at the tails in a stably miniature or oversized world. If the wearer moves only the lens slider and the render cameras do not follow, blur improves and scale stays wrong.

Studying it

Run a virtual–real size match: people adjust a virtual object until it matches a real object of known size in the hand, or they span the object’s width with both hands. Vary render-camera separation while holding physical lens spacing and sharpness fixed.

Independent variables: ratio of rendered IPD to true IPD, presence of a body referent (virtual hands, a known-size prop), match versus absolute estimate. Dependent variables: equivalent scale factor, distance-estimate bias, reports of “I grew / the world shrank”.

Screen out stereo-anomalous participants, or the size effect is confounded with “disparity was never in use”.

Where it stops holding

Under monocular viewing, or with the stereo path switched off, rendered IPD barely acts; size falls to angular size and familiar size. In an abstract space with no familiar objects and no visible body, absolute scale is already poorly estimated and baseline error is swallowed by that uncertainty. Video see-through adds another scale layer if the camera baseline ≠ the eyes — that knob is not the engine’s virtual IPD. Short lab matches miss whether, ten minutes later, people have re-anchored scale on their virtual hands.

Applying it

• Drive render-camera separation from the wearer’s IPD setting, as the same datum as the lens slider, not as two independent numbers.
• Offer a calibration that uses the wearer’s own hand or a real referent, not only a millimetre box labelled “IPD”.
• Keep the default in a personal profile that can be changed; do not bake a factory mean into the scene.
• How to check: stand a virtual cup next to a real cup and change only render separation, not the mesh. If the cup becomes a toy or a tank, this setting is wrong. Correct texture scale on a single asset is not a pass.

Related

• Same group: N1.10.1 Binocular disparity only provides useful depth at near distances · N1.10.3 Some people have no stereo vision and need monocular depth-cue fallback · N1.10.4 Reducing disparity eases the conflict at the cost of flattening space
• Nearby: N1.07 IPD and Fit · N4.08 Spatial Scale and Size Perception
• Search terms: rendered IPD · virtual IPD · Lilliputism