Pith. sign in

REVIEW 4 major objections 5 minor 31 references

SceneExpander claims that a feed-forward 3D reconstruction model can be adapted at test time, using anchor distillation on captured views and self-distillation on an inserted AI-generated view, to expand a captured 3D scene with misaligned

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 17:14 UTC pith:GYUQ3MTC

load-bearing objection Real task, sensible recipe, but the inserted view is never actually used as a constraint — the central claim overreaches. the 4 major comments →

arxiv 2603.27084 v3 pith:GYUQ3MTC submitted 2026-03-28 cs.CV

SceneExpander: Text-Guided 3D Scene Expansion via Free-Form View Insertion

Zijian He , Renjie Liu , Yihao Wang , Weizhi Zhong , Huan Yuan , Kun Gai , Guangrun Wang , Guanbin Li This is my paper
classification cs.CV
keywords scene expansionview insertiontest-time adaptationdistillationmulti-view reconstruction3D Gaussian splattingtext-guided generationworld building
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tackles a practical world-building workflow: starting from a real scene captured by multiple images, a user prompts an image generator to produce an additional view that extends the scene, and this view is inserted into the 3D reconstruction. The inserted view is usually not 3D-consistent with the captured geometry, so naively adding it corrupts the original reconstruction. The authors claim that test-time adaptation of a feed-forward reconstruction model, using two distillation signals—anchor distillation on captured views and self-distillation on the inserted view—can integrate the misaligned view while preserving the captured region. If correct, this would let creators iteratively expand real 3D scenes with text-controlled AI content without requiring perfectly 3D-consistent generation. The paper reports improved preservation (PSNR 21.21 vs 19.30) and expansion quality over baselines.

Core claim

The central claim is that a parametric feed-forward 3D reconstruction model can be adapted at test time—using only internal distillation signals—to absorb a generatively synthesized view that is geometrically inconsistent with the original multi-view evidence. The adaptation freezes an 'anchor teacher' to provide fixed camera/depth/normal predictions on the captured views, pulling the student back toward the pre-insertion model; simultaneously, an EMA teacher provides soft geometric targets on the inserted view, and the student distills toward those soft labels with a small weight, so the new content is gently explained rather than hard-fitted. A stochastic-restoration step periodically nudg

What carries the argument

The load-bearing object is the dual-teacher test-time adaptation loop built on a feed-forward reconstruction backbone named WorldMirror in the paper. An anchor teacher (frozen at pre-insertion weights) emits reference camera, depth, and normal maps for captured views; the student is penalized for deviating from them. A generation teacher, updated as an exponential moving average of the student, emits soft 'pseudo-geometry' targets for the inserted view; the student is lightly distilled toward these, so the inserted content is integrated as a weak constraint. Subset sampling of captured views, geometry-perturbation augmentation, and stochastic restoration complete the mechanism. The identity

Load-bearing premise

The method assumes the pre-trained reconstruction model's own view of the inserted image—iteratively refined through the EMA teacher—is a trustworthy guide to what the new content should look like in 3D, with no external ground truth or multi-view check. If that prior is uninformative or wrong, the expansion will be confidently wrong.

What would settle it

Take a captured multi-view scene with known camera poses and ground-truth geometry. Insert a real photograph of a different room (or a heavily displaced copy of a captured view) as the 'generated' view, so the misalignment is controlled and the correct expansion is known. If the adapted model renders the inserted content coherently from novel viewpoints and preserves the captured region, the central claim holds; if it produces ghosting, suppresses the inserted room, or drifts on the captured side, the claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Expanded 3D scenes can be produced from a single misaligned AI-generated view, without retraining the reconstruction model or requiring 3D-consistent generation.
  • The captured region is not merely tolerably preserved; on the paper's metrics it is better preserved than the backbone's own output (PSNR 19.30 → 21.21, SSIM 0.614 → 0.776), because anchor distillation actively stabilizes it.
  • The framework supports iterative world building: a user can chain multiple text-guided insertions, each extending the reconstruction with new content.
  • The ablations indicate that each component contributes: anchor distillation, subset sampling with augmentation, and stochastic restoration all improve the final trade-off.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The success of inserted-view self-distillation suggests a general recipe for integrating unreliable observations into a pretrained 3D model: trust the model's own iteratively refined predictions over the raw observation, and anchor against old evidence. The same principle could apply to updating radiance fields with noisy depth sensors or to online learning under distribution shift.
  • Because the only supervision for new content is the backbone's prior on a single image, the method's expansion quality is bounded by that prior. If the generative model produces a view with content the backbone has never seen, SceneExpander will likely imprint the backbone's hallucination—a limitation the paper acknowledges in its failure case.
  • A natural testable extension is long-horizon, multi-step expansion: the paper's stochastic restoration and EMA momentum would need tuning to prevent error accumulation over many sequential insertions; the paper lists this as future work.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces SceneExpander, a test-time adaptation method for integrating a text-guided, AI-generated inserted view into an existing multi-view 3D reconstruction. The method builds on a feed-forward reconstruction backbone (WorldMirror) and adapts it with two distillation losses: (i) anchor distillation on captured views against a frozen teacher to preserve the original scene, and (ii) inserted-view self-distillation on the generated view against an EMA teacher to 'accommodate' the new content. Additional components include subset sampling, geometry-perturbation augmentation, and stochastic restoration. Experiments on ETH scenes and online data report that SceneExpander improves preservation metrics (PSNR/SSIM/LPIPS) and expansion quality (GPT-IS, GPT-RQ, human scores) over baselines. The core claim is that this test-time adaptation robustly integrates a 3D-misaligned inserted view while keeping the captured region faithful.

Significance. The problem formulation—expanding a captured 3D scene by inserting a potentially inconsistent AI-generated view—is timely and practically relevant for iterative world building. The proposed anchor distillation is a sensible mechanism for preserving the known region, and the ablation study is systematic. However, the central mechanism for actually introducing the new content is under-specified and appears self-referential: the inserted-view self-distillation only distills the student toward its own EMA teacher, with no pixel-level or external geometric supervision from the inserted view. The evaluation is also weakened by using the same model family (GPT-5.2) both to generate the inserted views and to score the outputs, by manual curation of the inserted views, and by the absence of error bars or statistical tests on any headline number. If the reported improvements are real, the method is a useful step; but the current evidence does not conclusively show that the scene is actually expanded rather than merely stabilized.

major comments (4)
  1. [§3.3, Eqs. (2), (7), (8)] The inserted-view self-distillation signal is self-referential. The soft labels in Eq. (7) are produced by the EMA teacher θ_G, which by Eq. (2) is an exponential moving average of the student's own parameters. Eq. (8) then distills the student toward those soft labels. No term compares the model's output to the pixels of I_g or to any external geometric ground truth. The only information channel from I_g to the reconstruction is the backbone's forward pass on a batch containing I_g. Since the paper itself states that feed-forward models suppress or distort content when a misaligned view is added, the EMA teacher may simply preserve that suppression. The self-distillation is therefore more accurately described as a temporal-consistency regularizer than as an 'expansion constraint.' This undermines the load-bearing claim that the inserted view contributes new scene content. The authors sh
  2. [§4.1, App. C] The evaluation of expansion quality is circular. The same model, GPT-5.2, is used to generate the inserted views and to compute the GPT-IS and GPT-RQ scores. A VLM evaluator will likely favor outputs that resemble its own generation style, so the reported gains in GPT-IS/GPT-RQ may partly reflect self-consistency rather than true expansion quality. The human study (39 participants) is a partial mitigation, but it reports only averaged rank-derived scores without confidence intervals, inter-rater agreement, or significance tests. Tables 1 and 2 report single averages with no error bars, so the reader cannot judge whether the improvements (e.g., PSNR 21.206 vs 19.295) are statistically meaningful. Please report per-scene variance, confidence intervals, and significance tests, and consider using an independent evaluator (e.g., a different VLM or an absolute human scoring protocol).
  3. [§4.1, App. A] The manuscript states that the authors 'manually curate the generated candidate views and select those that best match the user instructions.' This introduces a selection bias: the evaluation is performed on views that are already likely to be visually plausible and integrable, which weakens the claim of robustness against 'free-form' and '3D-misaligned' insertions. The paper does not quantify the degree of misalignment (e.g., pose error, reprojection error, or occlusion). To support the central claim, the authors should either characterize the misalignment distribution of the curated views, or include results on uncurated generated candidates.
  4. [Tab. 2, Ablations] The ablation shows that adding inserted-view self-distillation (p=0.5) changes GPT-IS from 3.56 to 3.70 and GPT-RQ from 3.24 to 3.08 compared to the previous variant, but no error bars are given and the differences are small. This does not convincingly demonstrate that self-distillation introduces new content; it could be an artifact of improved rendering stability. A more informative ablation would decouple the presence of I_g in the input batch from the self-distillation loss—for example, comparing a variant that receives I_g but has λ_G=0 (no self-distillation) against the full model—and then measuring whether the rendered expanded region actually contains the requested object. The current table conflates stabilization and expansion.
minor comments (5)
  1. [Tab. 1] Typo: 'OminiVGGT' should be 'OmniVGGT'.
  2. [App. A] Typos: 'expanison' should be 'expansion'; 'onlince' should be 'online'.
  3. [Eq. (9)] The 'Restore' operation is not precisely defined. Please specify how the random subset of parameters is selected and how the rate r affects the restoration.
  4. [§4.1] The model name 'GPT-5.2' is unusual; if this is a specific model version, please cite it. If it is anonymized, state that clearly.
  5. [Fig. 6] Typo in the caption: 'inerted' should be 'inserted'.

Circularity Check

1 steps flagged

Self-distillation on the inserted view is self-referential by construction: L_gen targets the student's own EMA, so the expansion signal reduces to the backbone's prior rather than to I_g.

specific steps
  1. self definitional [Section 3.3, Eqs. (2), (6)-(8)]
    "Since θG is updated as an EMA of the student (Eq. (2)), the soft label evolves over time, gradually incorporating improved predictions from the adapted model. We distill the student toward the soft label via L(t) gen = dcam( ˆP (t) g , ¯P (t) g ) + βD dD( ˆD(t) g , ¯D(t) g ) + βN dN ( ˆN (t) g , ¯N (t) g ), (8)"

    The target in L_gen (Eq. 8) is f_{θ_G}(B)|_g, and by Eq. (2) θ_G = μθ_G + (1−μ)θ, i.e., an EMA of the student's own parameters. Therefore the 'inserted-view constraint' is a temporally smoothed version of the student's own prediction on the same batch; no loss term compares the reconstruction to I_g's pixels or to external 3D ground truth. At a fixed point θ_G = θ, L_gen = 0 identically, so the final expansion is the backbone's own prediction, not a constraint derived from I_g. The objective stabilizes/retains the model's prior rather than introducing content from the inserted view.

full rationale

The paper's framework is a test-time adaptation of a feed-forward reconstructor. Anchor distillation uses a frozen teacher on captured views and is externally validated by PSNR/SSIM/LPIPS against captured images, so the preservation claim rests on independent ground truth. The self-referential element is the inserted-view self-distillation: the soft label on I_g is produced by the EMA generation teacher, which is a moving average of the student itself (Eq. 2). Thus L_gen enforces consistency with the model's own past predictions; at convergence it is identically zero, so it cannot by itself inject new content from I_g beyond what the pre-trained backbone already predicts from that image. This is a genuine circularity in the expansion-signal mechanism, but it is partial: subset sampling, augmentation, anchor distillation, and stochastic restoration are external constraints, and the human H-IS/H-RQ study provides an independent check of expansion quality. The use of GPT-5.2 as both the inserted-view generator and the GPT-IS/GPT-RQ evaluator is a separate evaluation-chain self-reference that should be noted, but it is a validity caveat rather than a derivation-chain circularity. Overall, the central preserved-region claims are externally grounded; the expansion signal is self-referential by design, warranting a moderate score of 4.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 0 invented entities

The system rests on the pre-trained backbone's priors, the anchor teacher's predictions as preservation targets, and the evaluation setup (manual curation of generated views, VLM-as-judge). Hyperparameters include six loss/momentum weights, two probabilities, a sampling distribution, and several unstated values (µ, K, T). No invented entities: no new particles, forces, or dimensions are postulated.

free parameters (7)
  • anchor loss weights α_D, α_N = 0.2, 0.2
    Chosen constants weighting depth and normal distances in L_anchor (Eq. 5); no sensitivity analysis reported.
  • self-distillation weights β_D, β_N = 1.0, 1.0
    Weights for inserted-view soft targets in L_gen (Eq. 8).
  • objective weights λ_A, λ_G, λ_reg = 1.0, 5e-2, 1e-4
    Overall loss balancing (Eq. 10); λ_G small by design so the inserted view is a soft constraint.
  • insertion probability p = 0.5
    Probability of including I_g in each mini-batch (Eq. 3); the ablation compares p=0.5 versus p=1.0.
  • EMA momentum µ = unstated
    Controls soft-label refinement speed in Eq. (2); value never given in the text.
  • stochastic restoration rate r and interval K = r = 1e-3; K unstated
    Controls periodic partial reset toward θ_0 (Eq. 9); K never given.
  • subset size distribution = sizes in {n−1, n−2, n−3}
    Sampling scheme for captured-view subsets (Eq. 3); probabilities over the three sizes are not specified.
axioms (6)
  • domain assumption The backbone (WorldMirror) predicts informative geometry for a single 3D-misaligned inserted view I_g.
    L_gen (Eq. 8) distills the student toward its own EMA; no external 3D supervision for the new content exists. If the backbone's prior on I_g suppresses or distorts the inserted content, the loop stabilizes a wrong expansion. §3.3, Eqs. (7)-(8).
  • domain assumption The frozen anchor teacher's predictions on captured views are a valid preservation target.
    L_anchor (Eq. 5) anchors to the pre-trained model's own initial predictions (P̄, D̄, N̄), not ground-truth geometry; validity is only indirectly checked by image-space metrics on real images. §3.3, Eqs. (4)-(5).
  • domain assumption Manually curated GPT-5.2-generated views define a fair benchmark for expansion quality.
    Inserted views are synthesized by GPT-5.2 and then filtered by the authors to best match prompts (§4.1, App. A); the test set is a curated subset and all comparisons inherit that selection.
  • domain assumption GPT-5.2 VLM scores (GPT-IS, GPT-RQ) are human-aligned assessments of the outputs.
    Headline expansion metrics come from an LLM judge; the 39-participant study provides partial external validation, but the GPT numbers are used as primary evidence in Tabs. 1-2. App. C.
  • standard math Gradient-based optimization and EMA averaging behave as standard.
    No convergence or stability analysis is provided beyond the empirical ablations; the method assumes AdamW/EMA dynamics do what the ablation table suggests.
  • domain assumption The distance functions d_cam, d_D, d_N are well-defined and meaningful for the quantities they compare.
    d_D is described only as 'ℓ1 (or scale-invariant) depth distance' (§3.3), and d_cam is never specified — the losses' behavior depends on these choices.

pith-pipeline@v1.3.0-alltime-deepseek · 12509 in / 21090 out tokens · 202911 ms · 2026-08-02T17:14:44.097963+00:00 · methodology

0 comments
read the original abstract

World building with 3D scene representations is increasingly important for content creation, simulation, and interactive experiences, yet real workflows are inherently iterative: creators repeatedly extend existing scenes under user control. Motivated by this gap, we study text-guided 3D scene expansion via free-form view insertion. Starting from a real scene captured by multi-view images, a user specifies a text expansion intent, which a generative model materializes as an inserted view extending the scene coverage. Unlike simple object editing or style transfer within a fixed scene, the inserted view may be 3D-misaligned with the original reconstruction, introducing geometric shifts, hallucinated content, or view-dependent artifacts that disrupt global multi-view consistency. To address this challenge, we propose SceneExpander, which applies test-time adaptation to a parametric feed-forward 3D reconstruction model with two complementary distillation signals: anchor distillation stabilizes the captured scene using geometric cues from the captured views, while inserted-view self-distillation retains insertion-supported predictions to accommodate the misaligned view. Experiments on ETH scenes and online data demonstrate improved expansion behavior and reconstruction quality under misalignment.

Figures

Figures reproduced from arXiv: 2603.27084 by Guanbin Li, Guangrun Wang, Huan Yuan, Kun Gai, Renjie Liu, Weizhi Zhong, Yihao Wang, Zijian He.

Figure 1
Figure 1. Figure 1: Two examples of controllable 3D scene expansion via text-guided view insertion. Starting from a 3D reconstruction of cap￾tured views, the user requests an additional view beyond the observed region, which is synthesized and inserted to extend the scene. Yellow marks the view-consistent (captured) area, and red highlights newly in￾troduced content/objects. Abstract. World building with 3D scene representati… view at source ↗
Figure 2
Figure 2. Figure 2: Challenges in 3D scene expansion via free-form view insertion. (1) A 3D-misaligned inserted view can corrupt reconstruction quality in the captured region. (2) Newly introduced content must remain consistent under novel viewpoints. Red boxes highlight ghosting artifacts caused by conflicting 3D constraints from the inserted view. To answer this question, we propose SceneExpander, a framework for ex￾panding… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of SceneExpander. (1) A feed-forward reconstructor fθ pre￾dicts an initial 3D scene from captured views. (2) To accommodate a prompt￾generated inserted view Ig that may be 3D-misaligned, we adapt fθ at test time using two distillation losses: anchor distillation on captured views and inserted￾view self-distillation on Ig. (3) The adapted model produces an expanded 3D scene consistent with the capt… view at source ↗
Figure 4
Figure 4. Figure 4: Geometry-perturbation augmentation. We show four augmentation modes: identity (original), global affine warp, blockwise piecewise-affine warp (grid 3×3), and their combination (global+block), designed to mimic pose drift and local geometric inconsistencies. Anchor distillation on captured views. A central difficulty of view insertion is that the inserted image Ig can be geometrically inconsistent: directly… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of scene expansion under misaligned insertion. We show an indoor scene from the Online collection (left) and an outdoor scene from the ETH dataset (right). The first row presents the generated inserted view (purple box) together with a subset of captured multi-view images from the original scene (blue box). Each subsequent row shows renderings at novel camera poses for different meth… view at source ↗
Figure 7
Figure 7. Figure 7: More cases of SceneExpander. Left of the dashed line: input images, where the purple dashed box indicates the generated inserted view and the re￾maining images are captured views (the 2nd and 3rd cases show a representative subset of captured views for brevity). Right of the dashed line: novel views and corresponding depth maps rendered by camera navigation in the expanded scene. Input views Backbone predi… view at source ↗
Figure 6
Figure 6. Figure 6: A failure case of SceneExpander. The purple box shows the generated inserted view, and the red boxes highlight artifacts. SceneExpander alleviates misalignment on the two side pillars and the left stone pedestal, but still produces errors on the top crossbeam. Despite these improvements, SceneExpander mitigates geometric conflicts between the inserted view and the target scene via test-time adaptation, and… view at source ↗
Figure 8
Figure 8. Figure 8: User study interface. Participants rank videos A–E by insertion sat￾isfaction (H-IS) based on consistency with the edit goal, with the edited region highlighted in red. Options A–E include our method and four baselines, shown in a randomized order. The H-RQ interface follows the same format. D Broader Impact and Future Work Broader impact. SceneExpander supports a practical world-building workflow that exp… view at source ↗
Figure 9
Figure 9. Figure 9: Prompt for VLM-based evaluation on insertion satisfaction. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Prompt for VLM-based evaluation on rendering quality. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 6 linked inside Pith

  1. [1]

    In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference

    Cao, C., Yu, C., Liu, S., Wang, F., Xue, X., Fu, Y.: Mvgenmaster: Scaling multi- view generation from any image via 3d priors enhanced diffusion model. In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference. pp. 6045– 6056 (2025)

    2025

  2. [2]

    arXiv preprint arXiv:2511.23172 (2025)

    Chen, L., Li, R., Zhang, G., Wang, P., Zhang, L.: Fast multi-view consistent 3d editing with video priors. arXiv preprint arXiv:2511.23172 (2025)

    arXiv 2025

  3. [3]

    arXiv preprint arXiv:2509.26645 (2025)

    Chen, X., Chen, Y., Xiu, Y., Geiger, A., Chen, A.: Ttt3r: 3d reconstruction as test-time training. arXiv preprint arXiv:2509.26645 (2025)

    Pith/arXiv arXiv 2025

  4. [4]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Chen, Y., Chen, Z., Zhang, C., Wang, F., Yang, X., Wang, Y., Cai, Z., Yang, L., Liu, H., Lin, G.: Gaussianeditor: Swift and controllable 3d editing with gaussian splatting. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 21476–21485 (2024)

    2024

  5. [5]

    arXiv preprint arXiv:2509.21657 (2025)

    Dai, Y., Jiang, F., Wang, C., Xu, M., Qi, Y.: Fantasyworld: Geometry- consistent world modeling via unified video and 3d prediction. arXiv preprint arXiv:2509.21657 (2025)

    arXiv 2025

  6. [6]

    In: Proceedings of the IEEE/CVF interna- tional conference on computer vision

    Haque, A., Tancik, M., Efros, A.A., Holynski, A., Kanazawa, A.: Instruct-nerf2nerf: Editing 3d scenes with instructions. In: Proceedings of the IEEE/CVF interna- tional conference on computer vision. pp. 19740–19750 (2023)

    2023

  7. [7]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    He, Z., Ning, Y., Qin, Y., Wang, G., Yang, S., Lin, L., Li, G.: Vton 360: High- fidelity virtual try-on from any viewing direction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 26388–26398 (2025)

    2025

  8. [8]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    H¨ ollein, L., Cao, A., Owens, A., Johnson, J., Nießner, M.: Text2room: Extract- ing textured 3d meshes from 2d text-to-image models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 7909–7920 (2023)

    2023

  9. [9]

    arXiv preprint arXiv:2506.04225 (2025)

    Huang, T., Zheng, W., Wang, T., Liu, Y., Wang, Z., Wu, J., Jiang, J., Li, H., Lau, R.W., Zuo, W., Guo, C.: Voyager: Long-range and world-consistent video diffusion for explorable 3d scene generation. arXiv preprint arXiv:2506.04225 (2025)

    Pith/arXiv arXiv 2025

  10. [10]

    ACM Trans

    Kerbl, B., Kopanas, G., Leimk¨ uhler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph.42(4), 139–1 (2023) 16 Z. He et al

    2023

  11. [11]

    In: Proceedings of the Com- puter Vision and Pattern Recognition Conference

    Lee, D.I., Park, H., Seo, J., Park, E., Park, H., Baek, H.D., Shin, S., Kim, S., Kim, S.: Editsplat: Multi-view fusion and attention-guided optimization for view- consistent 3d scene editing with 3d gaussian splatting. In: Proceedings of the Com- puter Vision and Pattern Recognition Conference. pp. 11135–11145 (2025)

    2025

  12. [12]

    arXiv preprint arXiv:2510.10726 (2025)

    Liu, Y., Min, Z., Wang, Z., Wu, J., Wang, T., Yuan, Y., Luo, Y., Guo, C.: World- mirror: Universal 3d world reconstruction with any-prior prompting. arXiv preprint arXiv:2510.10726 (2025)

    arXiv 2025

  13. [13]

    arXiv preprint arXiv:1711.05101 (2017)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    Pith/arXiv arXiv 2017

  14. [14]

    arXiv preprint arXiv:2512.22096 (2025)

    Mao, X., Li, Z., Li, C., Xu, X., Ying, K., He, T., Pang, J., Qiao, Y., Zhang, K.: Yume-1.5: A text-controlled interactive world generation model. arXiv preprint arXiv:2512.22096 (2025)

    arXiv 2025

  15. [15]

    Commu- nications of the ACM65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021)

    2021

  16. [16]

    arXiv preprint arXiv:2511.10560 (2025)

    Peng, H., Li, H., Dai, Y., Lan, Y., Luo, Y., Qi, T., Zhang, Z., Zhan, Y., Zhang, J., Xu, W., et al.: Omnivggt: Omni-modality driven visual geometry grounded transformer. arXiv preprint arXiv:2511.10560 (2025)

    arXiv 2025

  17. [17]

    In: Proceedings of the SIGGRAPH Asia 2025 Conference Papers

    Schneider, M.A., H¨ ollein, L., Nießner, M.: Worldexplorer: Towards generating fully navigable 3d scenes. In: Proceedings of the SIGGRAPH Asia 2025 Conference Papers. pp. 1–11 (2025)

    2025

  18. [18]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 4104–4113 (2016)

    2016

  19. [19]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schops, T., Schonberger, J.L., Galliani, S., Sattler, T., Schindler, K., Pollefeys, M., Geiger, A.: A multi-view stereo benchmark with high-resolution images and multi-camera videos. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 3260–3269 (2017)

    2017

  20. [20]

    Vachha, C., Haque, A.: Instruct-gs2gs: Editing 3d gaussian splats with instructions (2024),https://instruct-gs2gs.github.io/

    2024

  21. [21]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wang, J., Chen, M., Karaev, N., Vedaldi, A., Rupprecht, C., Novotny, D.: Vggt: Visual geometry grounded transformer. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5294–5306 (2025)

    2025

  22. [22]

    In: European conference on computer vision

    Wang, Y., Yi, X., Wu, Z., Zhao, N., Chen, L., Zhang, H.: View-consistent 3d editing with gaussian splatting. In: European conference on computer vision. pp. 404–420. Springer (2024)

    2024

  23. [23]

    IEEE transactions on image processing 13(4), 600–612 (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing 13(4), 600–612 (2004)

    2004

  24. [24]

    arXiv preprint arXiv:2403.08733 (2024)

    Wu, J., Bian, J.W., Li, X., Wang, G., Reid, I., Torr, P., Prisacariu, V.A.: Gaussc- trl: multi-view consistent text-driven 3d gaussian splatting editing. arXiv preprint arXiv:2403.08733 (2024)

    Pith/arXiv arXiv 2024

  25. [25]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wu, T., Yang, G., Li, Z., Zhang, K., Liu, Z., Guibas, L., Lin, D., Wetzstein, G.: Gpt- 4v (ision) is a human-aligned evaluator for text-to-3d generation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 22227–22238 (2024)

    2024

  26. [26]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion

    Yang, X., Zuo, Y., Ramasinghe, S., Bazzani, L., Avraham, G., van den Hengel, A.: Viewfusion: Towards multi-view consistency via interpolated denoising. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. pp. 9870–9880 (2024) SceneExpander 17

    2024

  27. [27]

    IEEE Transactions on Visualization and Computer Graphics30(12), 7749–7762 (2024)

    Zhang, J., Li, X., Wan, Z., Wang, C., Liao, J.: Text2nerf: Text-driven 3d scene generation with neural radiance fields. IEEE Transactions on Visualization and Computer Graphics30(12), 7749–7762 (2024)

    2024

  28. [28]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 586–595 (2018)

    2018

  29. [29]

    arXiv preprint arXiv:2512.05354 (2025)

    Zheng, Y., Tan, H., Zhang, K., Wang, P., Guibas, L., Wetzstein, G., Yifan, W.: Splatpainter: Interactive authoring of 3d gaussians from 2d edits via test-time training. arXiv preprint arXiv:2512.05354 (2025)

    Pith/arXiv arXiv 2025

  30. [30]

    arXiv preprint arXiv:2401.14828 (2024)

    Zhuang, J., Kang, D., Cao, Y.P., Li, G., Lin, L., Shan, Y.: Tip-editor: An ac- curate 3d editor following both text-prompts and image-prompts. arXiv preprint arXiv:2401.14828 (2024)

    Pith/arXiv arXiv 2024

  31. [31]

    In: SIGGRAPH Asia 2023 conference papers

    Zhuang, J., Wang, C., Lin, L., Liu, L., Li, G.: Dreameditor: Text-driven 3d scene editing with neural fields. In: SIGGRAPH Asia 2023 conference papers. pp. 1–10 (2023) 18 Z. He et al. APPENDIX A Inserted View Generation Details We generate the inserted view on ETH dataset and onlince collections using a vision-language model conditioned on an exploration-...

    2023