Skip to content
Computational Pathology Reading Roadmap

Writing study: SceneSplat: Gaussian Splatting-based Scene Understanding with Vision-Language Pretraining

← ICCV 2025 oral writing corpus

Yue Li, Qi Ma, Runyi Yang, Huapeng Li, Mengjiao Ma, Bin Ren, Nikola Popovic, Nicu Sebe, Ender Konukoglu, Theo Gevers, Luc Van Gool, Martin R. Oswald, Danda Pani Paudel · Official ICCV 2025 oral record

Figure, table and section ledger

Editorial contribution type: Datasets, benchmarks & measurement (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF
12 physical pages
Before references / appendix
≈ 9.05 page-equivalents
Main figures / tables
6 / 9
Attached appendix figures / tables
0 / 0
Attached appendix span
No boundary detected; not assumed zero
Before references / appendix: ≈ 75.42%References / intervening material: ≈ 24.58%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.

Source-bound correction: The Table8 label and caption body are separated in extracted text; the p.8 raster confirms one table. Main caption inventory is Figures1–6 and Tables1–9.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level headingStarts on PDF pageEstimated spanShare of entire PDF
1. Introduction1≈ 0.86 pp.7.17%
2. Related Work2≈ 1.07 pp.8.92%
3. SceneSplat Dataset3≈ 0.45 pp.3.75%
4. Methodology4≈ 2.02 pp.16.83%
5. Experiments6≈ 2.56 pp.21.33%
6. Conclusion8≈ 0.2 pp.1.67%
7. Acknowledgment9≈ 1 pp.8.33%
References10≈ 2.95 pp.24.58%
All detected figure and table locations

Main text

Attached appendix

    No recognized caption labels.

    Argument and organization

    1. 1 Introduction; 2 Related Work

      Pairs the lack of native Gaussian semantic models with the lack of scene-scale training data.

    2. 3 SceneSplat Dataset; 4 Methodology

      Documents filtering and language-label collection before distinguishing vision-language from label-free pretraining.

    3. 5 Experiments; 6 Conclusion

      Separates zero-shot semantics, self-supervised transfer, data quality, label fidelity, scaling and runtime.

    Problem, claim and contribution

    Makes dataset and model co-dependence explicit instead of treating a large converted corpus as a background detail.

    Sections 1 and 3, physical pp.1–4 · Main paper

    Argument and section order

    Dataset construction precedes the model, and Algorithm 1 makes the 2D teacher-to-Gaussian label pathway inspectable before losses are introduced.

    Section 3 and Algorithm 1, pp.3–4 · Main paper

    Figures, captions and visual explanation

    A split overview gives the two pretraining regimes separate panels; the self-supervised panel labels teacher/student, EMA and four distinct loss paths.

    Figure 2, p.5 · Main paper

    Tables, comparisons and evidence

    The zero-shot table preserves training-source and relative-scene-count columns, foreground-only class exclusions and copied-baseline provenance; the SSL table retains the ScanNet++ performance drop.

    Tables 2–3, p.7 · Main paper

    Appendix and reproducibility

    Per-dataset processing, licenses, implementation and additional qualitative examples are assigned to the supplement; only main-text pointers were inspected.

    Sections 3,5 and 5.2, pp.3–4,6–7 · Main paper

    Clarity and writing conventions

    Dataset statistics distinguish approximately 9,000 raw scenes from 7,916 curated scenes, avoiding reliance on the rounded 7K name as an exact count.

    Table 1 and Section 3.2, pp.3–4 · Main paper

    What to reuse—and what not to infer

    Transferable lesson: For joint data-and-method papers, expose conversion quality, teacher-label provenance and downstream evaluation regimes before summarizing gains.

    Boundary: Zero-shot refers to downstream label evaluation, not absence of pretrained 2D supervision. PSNR–semantic-quality association is not a controlled causal intervention; Gaussian construction and label collection are not included in a single network forward-pass claim.

    Apply this lesson using the argument and evidence worksheet.

    Inspection record and sources

    Checked
    2026-10-07
    Reading status
    Writing-focused manuscript scan + selected visuals inspected
    Main text
    Writing-oriented scan across the main manuscript, physical PDF pp.1–8, covering framing, section progression, method explanations, captions, comparison protocols and concluding scope; reference boundary checked. Selected full-page rasters were inspected. Not exhaustive line-by-line reading, proof verification or experimental reproduction.
    Appendix / supplement
    Main-paper appendix/supplement references were checked. A separate supplementary source is linked in the inventory but was not retrieved or inspected; its contents are not included in the reading or visual-page counts.
    Visually inspected PDF pages
    5, 7

    This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.