A VR headset renders two views of the same scene from cameras separated by the interpupillary distance (IPD), ~63 mm on average. The correct method is off-axis (asymmetric-frustum) projection: both eye cameras stay parallel — no rotation — and only the projection is skewed sideways so the two frustums agree exactly on the whole convergence plane. For a world point at depth Z and horizontal position X, viewed from an eye offset by e, the projected screen position with a shared screen plane at the convergence distance F is:
screen = e + (X − e) · F / Z
At Z = F this reduces to screen = X for every eye offset — zero disparity everywhere on that plane, not just at the centre.
The common mistake is toe-in: keep each eye's frustum symmetric, but rotate ("verge") both cameras inward to aim at the convergence point, like real eyeballs. Toe-in only matches off-axis exactly at that one aim point. Off that point the two symmetric, rotated frustums disagree (the toe-in geometric error readout) — the 3D analogue of this is the vertical-parallax artifact that gives users eye-strain; this 2D top-down model shows its horizontal, in-plane sibling, keystone disagreement, directly.
Separately, VR architecture walkthroughs are often built at a model scale other than the real building's 1:1 size. Touring a model built at scale s with a camera rig still using a real human's 63 mm IPD makes disparity cues meant for a full-size room arrive at s× the distance they should — the brain reads that mismatch as the room itself being 1/s times its intended size. Combined with an IPD tuned incorrectly for the viewer, the two effects multiply into the familiar VR giant/miniature world illusion:
perceived scale ≈ modelScale × (assumed viewer IPD 63 mm / rig IPD)
- IPD slider — the rig's eye separation; wider IPD alone makes the world read smaller ("miniaturisation"), narrower reads bigger ("gigantism").
- Convergence slider — the depth plane with zero disparity ("the screen"); objects nearer show negative parallax, farther show positive parallax.
- Model scale slider — how large the architectural model is built relative to the real building it represents.
- Far/near-marker disparity — measured horizontal screen-space offset of the red wall marker (far) and blue table marker (near) between the two eye views, computed from the real off-axis or toe-in projection formula above.