Writing study: Self-Ensembling Gaussian Splatting for Few-Shot Novel View Synthesis

ICCV overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICCV 2025 oral writing corpus

Chen Zhao, Xuan Wang, Tong Zhang, Saqib Javed, Mathieu Salzmann · Official ICCV 2025 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Training, objectives & optimization (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

11 physical pages

Before references / appendix

≈ 8.05 page-equivalents

Main figures / tables

7 / 6

Attached appendix figures / tables

0 / 0

Attached appendix span

No boundary detected; not assumed zero

Before references / appendix: ≈ 73.18%References / intervening material: ≈ 26.82%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.

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 heading	Starts on PDF page	Estimated span	Share of entire PDF
1. Introduction	1	≈ 0.87 pp.	7.91%
2. Related Work	2	≈ 0.51 pp.	4.64%
3. Method	3	≈ 2.29 pp.	20.82%
4. Experiments	5	≈ 3.23 pp.	29.36%
5. Conclusion	8	≈ 0.27 pp.	2.45%
References	9	≈ 2.95 pp.	26.82%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Figure 3 · p. 4
• Figure 4 · p. 4
• Table 1 · p. 6
• Table 2 · p. 6
• Table 3 · p. 6
• Table 4 · p. 7
• Figure 5 · p. 7
• Figure 6 · p. 7
• Table 5 · p. 7
• Table 6 · p. 8
• Figure 7 · p. 8

Attached appendix

No recognized caption labels.

Argument and organization

1. 1 Introduction; 2 Related Work

   Motivates sparse-view overfitting and the cost of independently trained ensembles.

2. 3 Method

   Defines Gaussian rendering, uncertainty buffers, targeted perturbation and a second model that absorbs ensemble supervision.

3. 4 Experiments and analysis; 5 Conclusion

   Separates view-count benchmarks, training/inference costs and controls for perturbation diversity.

Problem, claim and contribution

Frames the problem as train/test divergence under too few views, using learning curves before introducing a regularizer.

Figure 2 and Section 1, physical pp.1–2 · Main paper

Argument and section order

Uncertainty computation is explained before perturbation selection and self-ensembling loss; the inference model is distinguished from the training pair.

Sections 3.3–3.4, pp.3–5 · Main paper

Figures, captions and visual explanation

The overview links render buffers to uncertainty maps, red perturbed Gaussians and the cross-model consistency loss; a separate FIFO-style buffer diagram explains temporal state.

Figures 3–4, p.4 · Main paper

Tables, comparisons and evidence

View-count tables retain metric directions and DTU background-mask rules; efficiency is explicitly a single fern scene while the model-count plots concern three-view LLFF.

Table 2, p.6; Figure 6 and Table 5, p.7 · Main paper

Appendix and reproducibility

Threshold choice, the set of overlapping pixels, split details and implementation settings are delegated to supplementary material; only their main-paper pointers were checked.

Sections 3.3 and 4.1, p.5 · Main paper

Clarity and writing conventions

The conclusion keeps unreliable perturbed models as a future-work concern rather than asserting uncertainty selection has eliminated all noisy supervision.

Section 5, p.8 · Main paper

What to reuse—and what not to infer

Transferable lesson: Draw stateful training buffers and identify the single inference model; separate scene-specific resource checks from dataset-wide accuracy claims.

Boundary: Rendering variance across training steps is an operational uncertainty proxy, not calibrated predictive uncertainty. Two trained models do not by themselves determine total cost: the fern comparison also changes Gaussian count substantially, so this one-scene comparison cannot establish uniform overhead or savings.

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 and the reference boundary were checked. No separate supplementary link appears in this dated CVF inventory; this does not establish that no supplementary material exists elsewhere.

Visually inspected PDF pages

4, 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.

• Official CVF open-access ICCV 2025 accepted-paper PDF, retrieved 7 October 2026; CVF states identical to accepted version except watermark; not assumed identical to initial submission · 11 pages