Panoramic footage is inherently 3DoF; translation cannot be added in post

What it is

A panoramic film pastes a full ring of pixels onto a sphere whose centre is the camera that shot it. The head can turn inside that centre; that is all omnidirectional / 360 video can give. Lean forward, sidestep, stand up, and near objects on the sphere never yield a new side — those rays were not recorded. Translation cannot be solved later from one equirectangular frame. That is missing viewpoints, not a codec failure.

Stereo panoramas (a sphere per eye) add a little disparity. They still cannot synthesise the third and fourth viewpoints walking needs.

Why it happens

A pinhole at one position and one orientation records one bundle of rays. Stitching orientations into a sphere fills angle, not position. A correct picture from a new location needs radiance along a new ray; surfaces hidden by foreground, and object flanks, never entered the lens, so any reprojection is invention.

At runtime a lean leaves two honest options: the sphere follows the head, near objects have no parallax, and the world feels painted on a visor; or the sphere is nailed to the room, and the image stretches under the wrong perspective as soon as the head moves. Neither is a six-degree-of-freedom scene. Light fields, multi-camera volumetric capture, or a neural radiance field recorded other rays — that is a different acquisition, not “upgrading the existing panorama”.

So on a position-tracked headset, panorama may follow rotation and must refuse or openly degrade translation. It cannot pretend the room is walkable.

Studying it

Let people look freely in a panorama, then additionally ask them to “step toward the object of interest / lean to see under the table”. Record whether they step, and whether they report perspective breakage or a presence break afterwards. Use a volumetric or real-time rendered take of the same scene as control.

Independent variables: acquisition (monoscopic sphere / stereo sphere / volumetric), whether physical walking is allowed, distance to the object of interest. Dependent variables: spontaneous translation count, reports of artifact or discomfort after a lean, the spatial-presence subscale of instruments such as IPQ.

A method note: playing a panorama on a six-degree-of-freedom headset does not give the footage six degrees of freedom. Label conditions by acquisition, not by headset model.

Where it stops holding

When everything of interest sits tens of metres away, head-range translational parallax is already weak and the missing motion is hard to notice. A handful of precomputed viewpoints can fake a small lean; at a half-metre step, holes and ghosts appear. Interactive content that depends on walking up to an object and grabbing it has no grabable geometry in a panorama. Live 360 is under the same single-viewpoint constraint as a finished file; what can be improved is resolution and stitch seams, not translation.

Applying it

• Design panoramas as films you can look around in: a seat, a viewing yaw, chapters — not as rooms you can enter.
• On a position-tracked device, make head translation an explicit refusal (a slight resistance, a caption “this title cannot be walked”) rather than letting the sphere drift without comment.
• If the recording must be walkable, change the shoot: volumetric, light field, or a CG rebuild. Do not promise “six degrees of freedom in post”.
• How to check: ask someone to lean or take a short step toward a near object in the film. If the object yields no new flank, or the sphere warps, translation was not recovered — that should be the expected result, and the product copy should say so.

Related

• Same group: N1.09.1 Yaw drifts over time and needs a recentre entry point · N1.09.2 The origin definition decides which side of the person content appears on · N1.09.4 Loss of positional tracking needs a degraded mode, not an interruption
• Nearby: N1.02 Degrees of Freedom · N1.01 Immersion and Presence
• Search terms: 360 video · omnidirectional video · motion parallax