ToggleInVR: VR Immersive Comparison
- ToggleInVR is an immersive video comparison method that displays one full-sphere view at a time, using toggling and a localized peek to enable sequential assessment.
- It preserves spatial registration by overlaying paired 360° videos in the same spherical coordinate system, reducing the cognitive load of recalling disparate scene details.
- Empirical results indicate that while ToggleInVR effectively detects fine-grained local differences under aligned conditions, its abrupt scene swaps may cause disorientation and increased memory demands.
ToggleInVR is an in-headset toggle technique for comparing two immersive 360° videos (IVs) when users cannot see both omnidirectional scenes side by side in VR. In the formulation studied in "To Slide or Not to Slide: Exploring Techniques for Comparing Immersive Videos," one IV is displayed on the entire inside of the viewing sphere at a time, the paired IVs are spatially overlapping in the same spherical display space, and the controller’s “B” button swaps the active full-sphere view; a localized peek reveal supplements this otherwise discrete comparison model (Wang et al., 22 Feb 2026). The technique is significant as a baseline superposition strategy for immersive comparison: it preserves viewpoint alignment and familiar state-of-the-art IV viewing conditions, yet exposes the memory, orientation, and workflow costs of comparing spherical media sequentially rather than simultaneously.
1. Definition and problem setting
ToggleInVR addresses a constraint specific to immersive video comparison: standard VR viewing ordinarily presents only one 360° scene at a time. The comparison task therefore becomes cognitively demanding because visual evidence is distributed across directions, and alternating between clips can require users to retain scene details in short-term memory while maintaining orientation within an egocentric spherical frame. The technique is the paper’s pure in-VR toggle condition, contrasted with SlideInVR’s hybrid sliding arrangement, ToggleIn2D’s desktop analogue, and SideBySideIn2D’s juxtaposition baseline (Wang et al., 22 Feb 2026).
The defining display rule is simple. ToggleInVR “displays a single IV on the entire sphere, similar to how users watch IVs in state-of-the-art systems.” The two videos are conceptually overlapping and spatially co-registered in the same spherical viewing frame, but only one is visible as the active full-sphere view. Because the viewing direction remains stable when the user toggles, the matching relation is spatial rather than side-by-side. The paper does not present a formal registration algorithm, but the design implies comparison in the same egocentric spherical coordinate system.
The scope of the technique is broader than perfectly synchronized same-camera captures. The study dataset included pairs from “the same environment and perspective,” “the same environment but different perspectives,” and “entirely different environments.” All clips were 30 seconds long and were presented for pairwise comparison. Playback controls were available across techniques. Temporal alignment was only partial and task-dependent; the paper does not report a specialized synchronization mechanism beyond playback controls and restarting videos.
A recurrent misconception is to treat ToggleInVR as a general VR comparison solution. The study instead positions it as a deliberately constrained baseline: one-at-a-time full-scene toggling with limited concurrent revelation, designed to expose the strengths and weaknesses of pure superposition in immersive media.
2. Interaction model and interface composition
The core interaction loop consists of normal 360° viewing, head- and body-based physical navigation, discrete full-scene toggling, and localized peeking. Users view one IV in the headset, orient themselves as in standard immersive playback, and press “B” to switch the entire sphere to the other IV. With overlapping IVs, peek reveals a small region of the other IV while the current view remains active. The paper specifies peek for the overlaid VR techniques as a small circular reveal measuring both vertically and horizontally, activated by pressing any trigger button; in ToggleInVR, this is the technique’s only concurrent-comparison affordance (Wang et al., 22 Feb 2026).
The interaction design preserves several elements from SlideInVR but with reduced scope. The minimap, progress bar, and spatially visualized ROI trajectory follow the same design as in SlideInVR. In that design, each minimap is an azimuthal equidistant projection attached above a controller and includes the current display area, a FoV indicator, a front-direction arrow (“F”), and an ROI indicator, together with a 3D progress bar and spatialized ROI trajectory. ToggleInVR inherits this family of tools, but only the current IV’s minimap is visible, and the projection always shows its full overview.
The hidden menu is correspondingly restricted. Only Restart Videos and Open/Close Panel are provided. No ROI-based advanced features are supported because ToggleInVR presents only one IV at a time and users are not able to reposition the IV. Unlike SlideInVR, the technique does not let users repartition spherical space, place ROIs side by side, partially overlap selected regions, slide display areas around the polar axis, or swap layered lunes.
This interaction model yields a distinctive balance. ToggleInVR preserves the familiar perceptual condition of full-sphere immersive viewing and keeps viewpoint alignment stable, but it does so by making comparison largely sequential. Peek mitigates that sequentiality only locally.
3. Rendering stack, platform, and experimental configuration
ToggleInVR shares its implementation stack with SlideInVR. The VR techniques were implemented in Unity $2022.3.16f1$ using the Meta XR Interaction SDK , and the system ran on a Meta Quest 3 with a vertical field of view and a horizontal field of view. Participants used a Quest 3 HMD and two controllers while seated on a wheeled office chair (Wang et al., 22 Feb 2026).
The rendering environment is a full 360° sphere. Unlike SlideInVR, the paper does not describe a separate custom shader pipeline for ToggleInVR beyond the fact that the videos are overlapping and one occupies the entire sphere at a time. It also does not report the video decoding method, streaming mechanism, frame-synchronization implementation, or latency measurements. No explicit mathematical formulas, state-transition diagrams, or algorithmic notation are given for ToggleInVR.
The empirical evaluation used a five-condition within-subjects design: SlideInVR, ToggleInVR, SlideIn2D, ToggleIn2D, and SideBySideIn2D. Twenty participants completed all five techniques in balanced Latin square order. Each condition began with a slide-based introduction and a tutorial of four practice tasks, followed by four experimental tasks in random order, each using a unique IV pair. Across the study there were 20 pairs of 30-second 360° clips spanning four task types: temporal occurrence differences , spatial-temporal motion comparison , local visual attribute differences , and global frame/quality differences . For –$2022.3.16f1$0, each IV had one predefined ROI tracked using a pre-trained open-source 360° tracking model; $2022.3.16f1$1 had no ROI tracking.
The dependent measures included trial accuracy, subjective helpfulness per task type, NASA-TLX workload ratings, UMUX-Lite usability, rankings and preferences, and qualitative strategy and workflow reports from questionnaires and semi-structured interviews. Because Likert and NASA-TLX ratings were non-normal, the study used Friedman tests with Wilcoxon signed-rank post hoc tests with Holm–Bonferroni correction. Accuracy, treated as binary with “I am not sure” counted as incorrect, was analyzed with mixed-effects logistic regression including participant as a random intercept and Technique and TaskType as within-subject fixed effects.
4. Empirical performance and task-specific utility
Quantitatively, ToggleInVR did not produce a significant accuracy advantage or disadvantage relative to the other techniques. The mixed-effects logistic regression revealed no significant main effect of Technique $2022.3.16f1$2, no significant main effect of TaskType $2022.3.16f1$3, and no significant Technique $2022.3.16f1$4 TaskType interaction $2022.3.16f1$5. ToggleInVR’s mean accuracy was $2022.3.16f1$6 with $2022.3.16f1$7 and $2022.3.16f1$8. The paper also reported no significant overall differences in perceived usability across techniques $2022.3.16f1$9; ToggleInVR’s mean UMUX-Lite score was 0 (Wang et al., 22 Feb 2026).
| Measure | ToggleInVR | Reported interpretation |
|---|---|---|
| Mean accuracy | 1 | No significant technique effect |
| Accuracy SD | 2 | 3 |
| UMUX-Lite | 4 | No significant overall usability difference |
Task-level helpfulness results were similarly restrained. For 5, there was an omnibus effect of Technique 6, but no post hoc comparison reached significance after Holm–Bonferroni correction 7. For 8–9, no significant helpfulness differences emerged 0. For workload, only physical demand differed significantly across techniques overall 1, but no technique-specific post hoc result was reported as significant for ToggleInVR.
The qualitative findings are more discriminating than the omnibus statistics. ToggleInVR was effective for detecting fine-grained local differences when ROIs were aligned in space, and it was particularly suitable for comparing short IVs captured from the same camera or for high-workload, detail-focused comparison tasks. It was therefore well matched to subtle shape, posture, or quality differences when the relevant evidence appeared in the same location or perspective. Peek was important in that niche because it enabled a form of concurrent local comparison without abandoning the full-sphere overlay logic.
By contrast, the technique became cumbersome when ROIs moved, when the compared content appeared in different locations, or when scenes differed substantially. In those cases, the user could not maintain simultaneous awareness of both motion patterns and instead had to inspect each video sequentially.
5. Perceptual trade-offs, limitations, and failure modes
ToggleInVR’s principal limitation is structural: it preserves spatial alignment but imposes a heavy burden on short-term visual memory. That burden increases when ROIs are not in the same viewing direction, because users may need to rotate between toggles. The paper repeatedly characterizes ToggleInVR, together with ToggleIn2D, as pushing participants toward sequential viewing rather than concurrent comparison (Wang et al., 22 Feb 2026).
A second limitation is abrupt scene replacement. The paper does not mention transition animation or cross-fading when pressing “B”; the full-scene swap appears abrupt. Participants reported disorientation during full-scene toggling, and some described dizziness. This is an important distinction from SlideInVR: although ToggleInVR nominally preserves viewpoint alignment, it can still disrupt spatial continuity because the entire immersive context changes at once.
A third limitation is restricted flexibility. Because ToggleInVR presents only one IV at a time, users cannot reposition scenes, construct side-by-side arrangements, or perform ROI-based spatial rearrangements. The hidden menu omits those functions accordingly. The technique is therefore narrower than sliding approaches and narrower than side-by-side for scenarios requiring simultaneous awareness of multiple anchors.
The method is also under-specified at several systems levels. The paper does not report latency for toggling or peeking, does not provide a custom synchronization mechanism, and does not publish a dedicated algorithm, shader equation, or state machine. These omissions are consequential for reproducibility at the implementation level, even though the interaction concept itself is clearly defined.
A plausible implication is that ToggleInVR should be read less as a universal immersive comparison paradigm than as a sharply delimited operating point: full-sphere binary alternation, augmented by a small local reveal, useful when alignment is strong and comparison targets are local.
6. Relation to broader VR switching and toggling research
Within immersive-video research, ToggleInVR is part of a larger argument that no single comparison method is sufficient across all tasks. The same study concludes that users need flexibility in switching between comparison approaches, with SlideInVR and SlideIn2D emerging as the most flexible and favorite methods, while ToggleInVR occupies a narrower but still useful role for short, local, same-viewpoint comparisons (Wang et al., 22 Feb 2026).
Outside that specific literature, the exact name "ToggleInVR" is not standardized. Several other VR systems investigate related but distinct state-switching problems. Rear-interruption handling research studies front-surfaced rear passthrough modes such as FRPP, RPW, and RPAR rather than immersive-video comparison (Guo et al., 20 Mar 2025). World-switching research studies portals and worlds-in-miniature for efficient transitions between virtual environments (Gottsacker et al., 1 Feb 2026). Locomotion research studies undo-actions that restore previous position and, in some variants, orientation (Müller et al., 2023). Optical-cloaking research studies switchable VR, AR, and MR within a common opaque-headset pipeline (Choi, 2018). Palm-based window-management research studies body-centric switching, closure, positioning, and scaling of windows in VR (Wang et al., 13 Aug 2025). The JUNO detector-visualization system, although it does not use the literal label, implements analogous controller-driven toggles for detector subsystems, event-display panels, photon-trail modes, and simulation-versus-data views (Huang et al., 19 Sep 2025).
This suggests that ToggleInVR is best understood as one member of a broader VR design family organized around discrete state changes, partial reveals, and transition aids, but specialized to the comparison of paired immersive videos. Its defining characteristics are not general “VR toggling” in the abstract, but full-sphere alternation, stable spherical viewpoint alignment, and a localized peek mechanism for limited concurrent inspection.