{"id":"5538a331-748a-4bd8-b22d-6c5d3c019790","arxiv_id":"2412.06257","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"3D Gaussian Splatting research relevant to Extended Reality is organized into a five-part taxonomy with suggested future directions.","lead":"This paper surveys 272 papers on 3D Gaussian Splatting and maps which innovations are relevant to Extended Reality, proposing a taxonomy and future research directions. It is useful as a roadmap for researchers looking to apply fast 3D rendering to VR, AR, and MR.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Survey's core 'underexplored' claim rests on an unreported classification of XR relevance; the 152/272 count and 'three representative works' are not independently checkable.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: the corpus selection and XR-relevance classification are not transparent, which undermines the central quantitative claim. I agree with the conditional verdict because the paper's taxonomy and roadmap can still be useful even if the exact counts are imperfect; the issue is reproducibility, not a demonstrated falsehood. An honest stress-test does not manufacture a stronger objection. The concrete test I propose would settle whether the concern actually lands: if the authors release the data and an independent replication reproduces the 152/272 split and the short list of XR-specific works, the central claim is supported; if not, the 'underexplored' narrative would need to be softened. Since the reader already conditioned acceptance on this fix, no verdict change is needed.","tokens_in":12259,"tokens_out":2706,"duration_ms":29296,"concrete_test":"Require the authors to release the full list of 272 papers with per-paper labels (XR mention, XR-specific, none) and the exact keyword set and search procedure. Then perform an independent replication: query Semantic Scholar/arXiv for works citing Kerbl et al. 2023 up to October 2024, apply the same (once disclosed) keyword filter, and compare the resulting 152/272 ratio and the set of XR-specific papers. If the ratio differs by more than 10%, or if an independent annotator following a written protocol achieves Cohen's kappa below 0.6 on a random 50-paper subsample, the 'underexplored' claim is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim, that direct 3DGS application in XR remains underexplored, depends on two undocumented judgments: which 272 papers were selected, and how each paper was classified as 'explicitly mention XR', 'superficially mention XR', or 'specifically target XR'. Section I reports '152 out of 272' without giving the search queries, databases, inclusion criteria, or a release of the paper list. Section II-A identifies only 'three representative works' (VR-GS, DualGS, RGCA) as specifically targeting XR, but 'representative' is not defined operationally, and no evidence is shown that other works claiming AR/VR demonstrations were excluded. Figure 2's taxonomy assigns papers to XR-relevant categories without demonstrating that each cited paper either mentions XR or evaluates in an XR setting; for example, [24] (StopThePop) is listed under Low-Latency Streaming but is a general rendering optimization with no explicit XR evaluation. If the classification threshold is loose or inconsistent, both the 152/272 statistic and the 'underexplored' conclusion could shift materially. Because the survey's novelty and roadmap are motivated by this gap, the lack of a reproducible screening protocol is the single most load-bearing weakness. This is a falsifiability/reproducibility problem, not an identified error, but it prevents a reader from checking the paper's headline quantitative claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a survey/roadmap at the intersection of 3D Gaussian Splatting (3DGS) and Extended Reality (XR). The authors state that they reviewed 272 publicly available 3DGS papers, that 152 explicitly mention XR-related concepts, and that only three works (VR-GS, DualGS, RGCA) specifically target XR. They propose a five-part taxonomy of 3DGS innovations relevant to XR: 3D Content Creation, Rendering & Visualization, Interaction & Manipulation, System Optimization & Efficiency, and Specialized Applications. They then discuss future research directions including mesh modeling, dynamic scene representation, open-world scene understanding, hand tracking, passthrough capabilities, and immersive visualization. The central claim is that direct application of 3DGS in XR is notably underexplored, despite the technology's promise.","tokens_in":12509,"tokens_out":2859,"duration_ms":28311,"significance":"The paper addresses a timely and useful topic: no prior survey focuses specifically on the 3DGS-XR intersection, and the authors provide a reasonable qualitative organization of relevant innovations. The proposed taxonomy and future-directions list could serve as a starting point for researchers entering this area. The paper also honestly acknowledges in Section V that some works may have been omitted. However, the quantitative claims about the corpus (272 papers, 152 XR-mentioning, three specifically targeting XR) are not reproducible from the manuscript as written, and the classification decisions are not documented. Because the survey's novelty and roadmap are motivated by the claimed gap, this methodology gap is load-bearing. The paper would be significantly strengthened by a transparent screening protocol, a released paper list, and explicit criteria for the 'specifically targeting XR' category.","major_comments":[{"comment":"The claim that the survey is based on '272 publicly available papers' and that '152 out of 272' explicitly mention XR concepts is not accompanied by any description of the search queries, databases, time window, inclusion/exclusion criteria, or paper list. Since the paper's central thesis, that XR applications of 3DGS are underexplored, depends directly on these counts and on how XR relevance was judged, the methodology must be reported or the quantitative claims must be softened. Please provide a reproducible protocol (e.g., an appendix with the corpus, keyword list, screening procedure, and ideally inter-annotator agreement) or reframe the argument as a qualitative observation.","section":"Section I, third paragraph"},{"comment":"The statement 'we have identified three representative works that specifically leverage 3DGS to advance XR research and development' lacks an operational definition of 'representative' and 'specifically target XR'. It is therefore not checkable whether other works that discuss AR/VR demonstrations or XR-related hardware, such as those cited elsewhere in the taxonomy, were considered and excluded. Please define the inclusion criteria for this category and provide evidence that the selection is not arbitrary; alternatively, present the three works as illustrative examples rather than as an exhaustive enumeration.","section":"Section II-A"},{"comment":"The taxonomy assigns individual papers to XR-relevant categories without demonstrating that each cited paper either mentions XR or evaluates its method in an XR setting. For example, reference [24] (StopThePop) is listed under 'Low-Latency Streaming' and 'High-Performance Rendering', but the paper is a general sorting optimization for Gaussian splatting and does not appear to include XR-specific evaluation. If some entries are included because the authors believe they have 'potential' for XR, that should be stated explicitly and distinguished from work that actually demonstrates XR use. Without a consistent mapping rule, the taxonomy's validity and the 'underexplored' conclusion are difficult to assess.","section":"Figure 2 and Section III"}],"minor_comments":[{"comment":"The word cloud is said to be 'extracted from recent 3DGS literature that references XR-related keywords', but no details are given about the keyword list, text sources, stop-word removal, or frequency normalization. Please describe the generation process or remove the figure.","section":"Section I, Figure 1"},{"comment":"There is a typo: 'V olumetric videos' should be 'Volumetric videos'.","section":"Section II-A, DualGS paragraph"},{"comment":"The example 'GS-VTON [95] enables VR content personalization based on user preferences' appears to mischaracterize the paper, which is a virtual try-on method; it is not clear that it targets VR. Additionally, references [93] and [94] are point cloud completion/surface reconstruction works, not 3DGS papers, and citing them in support of 'mesh modeling' needs clarification.","section":"Section IV, Mesh Modeling"},{"comment":"The statement 'we may have omitted some works' in the conclusion is appropriate but sits in tension with the earlier claim of a 'comprehensive review'. Please state the limitations of the search procedure earlier in the paper, e.g., in the introduction or a dedicated methodology paragraph.","section":"Section V"},{"comment":"The reference list has inconsistent formatting: some entries lack venue details (e.g., [10], [51], [55]), some use 'arXiv preprint' without an identifier, and the author lists contain occasional spacing issues (e.g., 'Y' instead of 'Y.' in several entries). A careful proofread is needed.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is more of a position/roadmap piece than a systematic survey, and the journal may wish to consider whether the quantitative claims are necessary to its contribution. I believe the central qualitative claim is plausible and the roadmap is useful, so a revision that either adds a reproducible methodology or explicitly reframes the claims as qualitative would bring it to an acceptable standard. I would not support rejection on the current evidence, but the load-bearing methodology gap must be addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nRead the 3DGS-XR survey. The short version: it's a decent roadmap with a legitimate niche, but the headline number (152/272) is the softest part of the paper and shouldn't be quoted without the authors releasing their screening data.\n\nWhat's actually new: the XR-focused taxonomy. Prior 3DGS surveys are general; this one organizes around five areas (content creation, rendering, interaction, system optimization, specialized applications) with a clear eye to what XR requires—latency, streaming, mobile hardware, interactivity. The writing is crisp, the taxonomy is sensible, and the choice to highlight VR-GS, DualGS, and RGCA as dedicated XR systems is defensible. I also like the future directions: mesh modeling, dynamic scenes, open-world understanding, hand tracking, passthrough, immersive visualization. Nothing groundbreaking, but it gives a newcomer a structured map.\n\nThe soft spots are real but not fatal. The central claim—that direct 3DGS application in XR is underexplored—rests on a screening of 272 papers where 152 'explicitly mention XR concepts.' The authors don't report the search queries, databases, inclusion criteria, or the paper list, and they don't define 'superficially mention' vs. 'specifically target.' The stress-test note flags StopThePop [24] being listed under Low-Latency Streaming without an XR evaluation; I checked, and it's a general rendering optimization that mentions XR only in passing if at all. So the classification is loose. But the qualitative conclusion—only a handful of systems truly built for XR—is probably right. The problem is the paper outsources its credibility to a number that's not checkable. They also admit in the conclusion that they may have omitted works, which undercuts the 'comprehensive review' framing.\n\nThat said, this is a survey and roadmap, not a systematic review with clinical-grade methodology. The lack of a released protocol is a legitimate criticism, but I'd treat it as a request for an appendix, not grounds for rejection. The taxonomy itself is the contribution, and it's usable.\n\nWho's this for: graduate students or researchers entering 3DGS or XR who want a structured overview and a map of open problems. It's not a transformative paper, and it won't change how I think about the field. But as a roadmap it earns referee time.\n\nRecommendation: send it to peer review with a request that the authors either release the paper list and screening criteria or soften the quantitative claims to 'the majority of surveyed papers mention XR only in passing.' Conditional accept, minor revisions.\n\nBest.","headline":"Useful XR-focused 3DGS roadmap with a defensible taxonomy, but the 152/272 count is unverifiable and should be softened or backed by released screening data.","tokens_in":12987,"tokens_out":2226,"would_cite":false,"duration_ms":19805,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This survey argues that 3D Gaussian Splatting is a natural fit for extended reality, yet only three systems have actually built for XR, and it maps where the field should go next.","keywords":["3D Gaussian Splatting","Extended Reality","Virtual Reality","Augmented Reality","Neural Rendering","taxonomy","human avatar rendering"],"falsifier":"A reproducible literature search with declared databases, queries, and inclusion criteria that finds more than three 3DGS systems with explicit XR implementations and evaluations would falsify the underexploration claim. For instance, a forward search for 3DGS papers mentioning 'headset' or 'passthrough' and reporting end-to-end XR deployment would settle the count.","tokens_in":12057,"feed_emoji":"🥽","tokens_out":6358,"duration_ms":53287,"temperature":0.7,"pith_summary":"This paper reviews 272 published 3D Gaussian Splatting (3DGS) papers and finds that while 152 mention extended reality (XR), only three systems, VR-GS, DualGS, and RGCA, are explicitly built for XR environments. It argues that the direct application of 3DGS inside XR is underexplored, despite the technology's real-time performance and explicit 3D representation making it a natural fit for immersive systems. To organize the opportunity, the paper proposes a taxonomy of five innovation areas, 3D content creation, rendering and visualization, interaction and manipulation, system optimization and efficiency, and specialized applications, and lists six future research directions. A sympathetic reader would take this as a roadmap: 3DGS is ready, but the XR community has barely started to use it.","feed_headline":"Three 3D Gaussian Splatting systems target XR so far","feed_subtitle":"A 272-paper survey finds XR is mostly a buzzword; a new taxonomy maps where real progress can happen.","key_machinery":"The load-bearing object is the taxonomy itself, constructed by keyword-screening a corpus of 272 publicly available 3DGS papers and classifying their innovations into five XR-relevant areas: 3D content creation, rendering and visualization, interaction and manipulation, system optimization and efficiency, and specialized applications (including SLAM and medical XR). The taxonomy is what turns a collection of loosely related rendering papers into an argument that XR is a distinct, tractable application ground. Its companion device is the identification of the three direct XR systems, VR-GS, DualGS, and RGCA, each of which supplies a concrete technical template: physics-aware deformation via XPBD, joint/skin Gaussian decoupling plus compression, and relightable appearance with explicit eye and hair models.","core_discovery":"The paper's central claim is that the intersection of 3DGS and XR is not merely nascent but underexplored: although a majority of 3DGS papers reference XR concepts, almost none demonstrate 3DGS functioning inside an XR system. The authors identify exactly three exceptions, VR-GS (physics-aware interactive editing in VR), DualGS (compressed volumetric human performance for XR), and RGCA (relightable codec avatars with gaze control), and treat these as existence proofs for what a dedicated 3DGS-XR system looks like. From a screened corpus of 272 papers, the authors build a five-category taxonomy of 3DGS innovations they consider relevant to XR, and they propose mesh modeling, dynamic scene representation, open-world scene understanding, hand tracking, passthrough capabilities, and immersive visualization as the most promising forward directions.","pith_inferences":["A direct extension of the paper's method would be to publish the screening protocol and corpus so the 'only three' count can be independently verified and updated.","If the underexploration claim holds, then a standardized benchmark measuring latency, render quality, and interaction responsiveness of 3DGS inside headset pipelines would be the logical next step, though the paper does not propose one.","The three demonstrated systems all center on human bodies or heads, suggesting that telepresence and avatar applications will drive early 3DGS adoption in XR faster than other domains.","The taxonomy's separation of interaction and manipulation from rendering points to a deeper, implicit claim: XR needs controllable geometry and physics, not just photorealism, which is why explicit 3DGS may fit XR better than implicit neural fields."],"forward_implications":["3DGS research that targets XR should focus on system-level integration rather than generic rendering, since the paper's screening shows that is the unoccupied space.","The three direct XR systems, VR-GS, DualGS, and RGCA, provide concrete templates for physics-aware editing, compressed volumetric video, and relightable avatars that future XR work can build on.","The proposed future directions, such as passthrough reconstruction, open-world scene understanding, hand tracking, and immersive data visualization, define concrete areas where 3DGS innovation is most likely to pay off for XR.","The taxonomy implies that progress in XR will depend on advances across all five innovation areas together, not on rendering alone."],"supporting_citations":[{"why":"The original 3D Gaussian Splatting method whose real-time explicit representation is the technology the survey tracks.","marker":"[5]"},{"why":"VR-GS, one of the three direct XR systems; supplies the physics-aware interactive editing template for the underexploration argument.","marker":"[13]"},{"why":"DualGS, one of the three direct XR systems; supplies the compressed volumetric human performance example.","marker":"[12]"},{"why":"RGCA, one of the three direct XR systems; supplies the relightable avatar and gaze-control example.","marker":"[14]"}],"fun_headline_variants":["Only 3 of 272 3DGS papers actually enter XR","3DGS meets XR: survey finds only 3 working demos","XR's 3DGS promise remains mostly unrealized: 3 exceptions","A 272-paper survey shows 3DGS-XR is still a rare overlap","Three 3DGS systems dare to tackle XR; 269 stay in theory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The survey's count of 'only three' direct XR systems rests on the authors' unpublished screening of 272 papers; if the search queries or inclusion criteria change, the number of XR-specific systems could change materially.","fun_headline_variants_meta":{"raw":{"variants":["Only 3 of 272 3DGS papers actually enter XR","3DGS meets XR: survey finds only 3 working demos","XR's 3DGS promise remains mostly unrealized: 3 exceptions","A 272-paper survey shows 3DGS-XR is still a rare overlap","Three 3DGS systems dare to tackle XR; 269 stay in theory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000831,"raw_usage":{"total_tokens":3619,"prompt_tokens":926,"completion_tokens":2693,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":542,"completion_tokens_details":{"reasoning_tokens":2587}},"tokens_in":542,"tokens_out":2693,"duration_ms":17298,"temperature":1.0,"reasoning_tokens":2587,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T19:50:36.467489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A reproducible literature search with declared databases, queries, and inclusion criteria that finds more than three 3DGS systems with explicit XR implementations and evaluations would falsify the underexploration claim. For instance, a forward search for 3DGS papers mentioning 'headset' or 'passthrough' and reporting end-to-end XR deployment would settle the count.","supporting_citations":[],"review_version":1}