SideBySideIn2D for Immersive Video Comparison

• SideBySideIn2D is a pure juxtaposition technique that displays two monoscopic 360° videos side by side, each with independent navigation and no overlay.
• It reduces memory load and mitigates attentional tunneling by keeping both video contexts continuously visible during dynamic scene comparisons.
• The method employs independent pan/tilt controls, rectilinear view sampling, and flexible timeline navigation to support temporal and spatial analyses.

Searching arXiv for the exact topic and source paper to ground the article. arxiv_search(query="^^^^^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^^^^^"To Slide or Not to Slide: Exploring Techniques for Comparing Immersive Videos\"^^^^^^^^2 OR \2^^^^^^^^"SideBySideIn2D\"", max_results=5) SideBySideIn2D is a 2D comparison technique for two monoscopic 36^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^-degree immersive videos (IVs) that supports concurrent viewing with independent navigation in each pane. In the comparative framework of "To Slide or Not to Slide: Exploring Techniques for Comparing Immersive Videos," it is the pure juxtaposition technique in the 2D suite: the canvas is split into two equal views, there is no overlay, and both video contexts remain continuously visible. The technique was designed as the simplest, most familiar baseline for continuous side-by-side viewing when both contexts should remain visible at once, with the stated goals of reducing memory load and minimizing tool complexity (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

^^^^2 OR \2^^^^. Definition and conceptual role

SideBySideIn2D supports concurrent viewing of two 36^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^-degree videos with independent navigation in each pane. It prioritizes stable juxtaposition to avoid reliance on short-term memory and to mitigate the “fear of missing something” when comparing dynamic scenes. Conceptually, it instantiates Gleicher’s juxtaposition paradigm for immersive videos: two separate views, no overlay, and independent control per view (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

Its position within the 2D comparison suite is defined by contrast with two other techniques. SlideIn2D embodies the “sliding” metaphor and hybridizes juxtaposition and superposition, while ToggleIn2D is pure superposition with a single shared viewport and a Toggle button to switch which video is visible. SideBySideIn2D, by contrast, is pure juxtaposition: fixed two-pane split, continuous visibility of both videos, and no overlay.

This organization yields a specific cognitive profile. SideBySideIn2D aims to minimize reliance on visual memory and reduce attentional tunneling by keeping both contexts present. The paper associates this profile particularly with dynamic ROIs, differing viewpoints, temporal counting, and longer sequences, where persistent dual context is beneficial (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

2. Interface layout and interaction model

The interface uses a canvas split into two equal views, with the left pane assigned to Video A and the right pane assigned to Video B. Each view uses a rectilinear perspective viewport with fixed vertical FoV and fixed horizontal FoV. The reported parameters are PRESERVED_PLACEHOLDER_^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^ and PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^ for each half-canvas, determined by the pane’s aspect ratio in Three.js. In the general case with a perspective camera,

$hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$

where $W/H$ is the pane’s aspect ratio; under the half-split layout, this yields approximately $59^\circ$ (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

Each pane supports independent physical navigation. Horizontal dragging changes yaw, and vertical dragging changes pitch. A Reset button returns each pane to a default front-facing view. A ROIs SxS control automatically centers each predefined ROI within its own pane, placing them side by side to reduce search and setup time before comparison.

A minimap is provided per pane. Each minimap uses an azimuthal equidistant projection and shows an IV overview, the current ROI location, and a gradient-colored ROI trajectory over time, encoded from red to green, for spatial-temporal awareness. Temporal navigation is separated by video: controls below the canvas allow seeking via a progress bar, play/pause, rewind/fast-forward by $\pm 5$ s, inspection of current and total duration, and simultaneous restart of both videos through “Replay Videos.” The paper emphasizes that there is no overlay in SideBySideIn2D; peek and ROIs Overlay are not supported, in order to keep interaction simple and avoid cognitive overhead (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

3. Rendering pipeline and implementation characteristics

The implementation uses Vite and React for the user interface, and Three.js/WebGL for rendering. Each pane uses a perspective camera viewing the equirectangular video texture mapped to a spherical surface; navigation changes camera yaw and pitch so that the correct rectilinear viewport is sampled (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

The source videos are standard equirectangular immersive videos. The user-facing viewports render rectilinear perspective subviews, while the minimaps render azimuthal equidistant projections. This separation of source projection and display projection is central to the technique’s operation: the immersive source remains intact, but the comparison task is carried out through two simultaneously available rectilinear subviews.

Synchronization is deliberately flexible rather than rigid. The player provides independent time control per video, so both panes can be played in parallel or scrubbed separately. This supports both simultaneous viewing and sequential analysis. “Replay Videos” restarts both in sync from the beginning. The absence of overlays reduces fragment blending cost relative to SlideIn2D. The reported rendering load—two rectilinear subviews and two minimaps—remained within typical WebGL desktop budgets; on the study apparatus, a 28-inch monitor, no performance issues were observed (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

4. Experimental evaluation framework

The evaluation used a within-subjects design in which $N=20$ participants used all five comparison techniques: SlideInVR, ToggleInVR, SlideIn2D, ToggleIn2D, and SideBySideIn2D. The participant pool comprised 8 female and ^^^^2 OR \22^^^^ male participants aged ^^^^2 OR \2^^^^8–34, screened for prior VR/IV viewing experience. Most had at least one year of VR experience, and most watched IVs rarely. For 2D techniques, the apparatus was a 28-inch monitor with keyboard and mouse; VR techniques used Meta Quest 3 (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

The task set comprised four task types derived from visualization and video-comparison literature. T^^^^2 OR \2^^^^ concerned temporal occurrence differences, such as what appears earlier, frequency, or duration. T2 concerned spatial-temporal motion, such as moving toward or away, motion speed, and path shape. T3 concerned local visual differences, such as size, color diversity, and posture. T4 concerned global frame or quality differences, such as distortions, blurring, stitches, consistency, and same location.

The stimuli consisted of 2^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^ pairs of 3^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^-second 36^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^-degree videos, mostly natural YouTube clips, with two simulated clips for controlled quality artifacts. For T^^^^2 OR \2^^^^–T3, each immersive video contained one predefined ROI. ROIs were tracked with a pre-trained 36^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^° tracking model, and their trajectories were visualized in the minimaps. No ROI tracking was used in T4.

Each technique session included a tutorial followed by four randomized experimental tasks. The recorded measures were task accuracy, post-questionnaires with NASA-TLX, UMUX-Lite, custom Likert ratings on helpfulness, navigation, and effectiveness, and interviews about strategies and preferences. Likert and NASA-TLX ratings were analyzed with Friedman tests and Wilcoxon signed-rank post-hoc tests with Holm–Bonferroni correction. Trial-level accuracy was analyzed with mixed-effects logistic regression with participant random intercept. The significance threshold was $p < .05$ (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

5. Quantitative and qualitative findings

The quantitative results place SideBySideIn2D in a distinctive but not uniformly dominant position. For helpfulness ratings by task type, only T^^^^2 OR \2^^^^ showed an omnibus effect of Technique, with Friedman $p = 0.017$; post-hoc pairwise tests did not reach significance after Holm–Bonferroni correction ($p = 0.064$). T2–T4 showed no significant differences across techniques, with PRESERVED_PLACEHOLDER_^^^^2 OR \2(Wang et al., 22 Feb 2026) OR \2^^^^. For NASA-TLX, only Physical Demand showed a significant omnibus effect (PRESERVED_PLACEHOLDER_^^^^2 OR \2 OR \2^^^^), but pairwise tests involving SideBySideIn2D did not reach significance; descriptively, VR techniques showed higher physical demand than 2D techniques (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

For UMUX-Lite, the omnibus Friedman test gave PRESERVED_PLACEHOLDER_^^^^2 OR \22^^^^, not significant. Descriptively, SideBySideIn2D had the highest mean usability score, reported as 76.53, and it was significantly preferred over ToggleIn2D, which had a mean of 66.23; the Wilcoxon comparison SideBySideIn2D vs ToggleIn2D yielded PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^3. In preference rankings, the overall Friedman test showed a significant effect (PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^4), and SideBySideIn2D was significantly preferred over ToggleIn2D with Wilcoxon PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^5.

Accuracy results showed no significant main effect of Technique (PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^6), TaskType (PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^7), or interaction (PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^8). The per-technique mean accuracies were PRESERVED_PLACEHOLDER_^^^^2 OR \2^^^^9 for SlideInVR, $hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^ for ToggleInVR, $hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$^^^^2 OR \2^^^^ for SlideIn2D, $hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$2 for ToggleIn2D, and $hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$3 for SideBySideIn2D.

The qualitative findings explain why SideBySideIn2D remained highly regarded despite the absence of accuracy gains. Participants described side-by-side viewing as natural and familiar, especially for temporal occurrence and dynamic ROIs. Continuous dual visibility was reported to reduce the need to memorize details and to alleviate fear of missing events. Participants often played both videos simultaneously when scanning for time-step differences or event sequences, then alternated gaze between panes. Minimap trajectories were used to compare paths and locate ROIs, especially when motion was complex or ROIs intermittently disappeared. The ROIs SxS function was praised as a “fix everything for you” convenience because it reduced manual navigation overhead (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

6. Comparative positioning, limitations, and usage guidance

The technique’s comparative strengths are task-dependent. SideBySideIn2D excels when ROIs are dynamic or appear in different locations or perspectives, when both videos should be played in parallel and scanned for temporal occurrence differences, and when the task is long or sequential enough that continuous dual context reduces memory burden. SlideIn2D excels when users must fluidly switch between side-by-side and overlay within one workflow, when dynamic reallocation of space is useful, and when ROIs Overlay accelerates subtle local comparisons while ROIs SxS supports temporal tracking. ToggleIn2D excels when ROIs are spatially aligned and differences are fine-grained, videos are short, and overlay-style flicker comparison reduces visual search and highlights pixel-level differences (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

A compact comparison of the 2D techniques clarifies these trade-offs.

Technique	Core comparison mode	Most suitable conditions
SideBySideIn2D	Pure juxtaposition	Dynamic ROIs, differing viewpoints, temporal occurrence, long comparisons
SlideIn2D	Hybrid juxtaposition and superposition	Mixed workflows needing space reallocation and overlay
ToggleIn2D	Pure superposition	Fine-grained local differences with aligned ROIs

The limitations of SideBySideIn2D follow directly from its design. Because there is no overlay, it is less effective for fine-grained local differences when ROIs are spatially aligned. The split layout yields a smaller per-pane horizontal FoV of $hFoV = 2 \cdot \arctan(\tan(vFoV/2) \cdot (W/H)),$4, which can constrain context; large motion may require more panning than in VR or in a wider single-pane view. Screen real estate can limit detail on smaller displays. The paper also notes that mental alignment across disparate perspectives still requires attentional switching, even though cognitive load is reduced relative to toggling.

The design recommendations reported for effective use are correspondingly specific: favor side-by-side when tasks involve dynamic ROIs, distributed viewpoints, or sequential event detection; provide independent controls per pane for both parallel and sequential viewing styles; include ROI-based centering to reduce search and setup time; incorporate minimaps with azimuthal equidistant projection and gradient time-encoded trajectories; and avoid unnecessary overlays or peeking in this mode to preserve simplicity. When finer-grained overlay is needed, the paper recommends an explicit switch to a sliding or overlay technique rather than overloading SideBySideIn2D (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).

Within the study’s evidence-backed conclusion, SideBySideIn2D is the most straightforward juxtaposition tool in the 2D suite. It keeps both videos visible, reduces memory burden, and supports parallel playback and scanning. Its measured usability was high, and it was significantly preferred over ToggleIn2D, but the broader findings also indicate that no single technique dominates across all comparison scenarios. This suggests that SideBySideIn2D is most appropriately understood as a stable, low-friction baseline for immersive-video comparison, particularly for dynamic and temporally extended tasks, rather than as a universal replacement for overlay-based or hybrid techniques (&&&^^^^2(Wang et al., 22 Feb 2026) OR \2^^^^&&&).