Writing study: Consistent 3D Surface Flow Model with Global State

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Antoine Guedon, Shu Nakamura, Nicolas Dufour, Jiahui Lei, Ko Nishino, Angjoo Kanazawa · Official NeurIPS 2026 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: Architecture & representation (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

22 physical pages

Before references / appendix

≈ 10.42 page-equivalents

Main figures / tables

6 / 5

Attached appendix figures / tables

1 / 0

Attached appendix span

≈ 4.91 page-equivalents

Before references / appendix: ≈ 47.36%References / intervening material: ≈ 30.32%Attached appendix: ≈ 22.32%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Top-level heading and reference/appendix boundaries checked against source PDF text. Bibliography titles/author initials, numbered contribution/property bullets, diagram steps, equations and TOC entries are not substantive headings. Section spans are page-coordinate estimates, not ink area or prose-only length. This is an author manuscript mirror, not a verified accepted-version template. Source layout uses physical single-column y/page-height, including equation-heavy pages.

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	2	≈ 1.32 pp.	6%
2 Method	3	≈ 2.4 pp.	10.91%
3 Experiments	5	≈ 3.92 pp.	17.82%
4 Related Work	9	≈ 1.04 pp.	4.73%
5 Conclusion	10	≈ 0.54 pp.	2.45%
6 Acknowledgments	11	≈ 0.11 pp.	0.5%
References	11	≈ 6.67 pp.	30.32%
A Related work	18	≈ 1.38 pp.	6.27%
B Architecture details	19	≈ 1.11 pp.	5.05%
C Training	20	≈ 0.77 pp.	3.5%
D Guidance	21	≈ 0.47 pp.	2.14%
E Evaluation protocol.	21	≈ 1.17 pp.	5.32%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 4
• Figure 4 · p. 6
• Table 1 · p. 7
• Table 2 · p. 8
• Figure 5 · p. 8
• Table 3 · p. 9
• Table 4 · p. 9
• Figure 6 · p. 10
• Table 5 · p. 10

Attached appendix

• Figure 7 · p. 20

Argument and organization

1. Global state motivation

   Connects viewpoint redundancy to a fixed-size scene representation.

2. Encoder, decoder and guidance

   Explains the three interventions in the same order as the central architecture figure.

3. Quality, resolution and ablations

   Tests shared-state reconstruction under controlled view counts and separates native from generated reference surfaces.

Problem, claim and contribution

The argument begins with geometry being independent of viewpoint, then derives why a shared latent should not grow with the number of views. This links the representation choice to the problem rather than introducing a module without a need.

Introduction; Section 2 · Main paper

Argument and section order

Encoder, flow decoder and guidance become explicit subsections matching the three-part method overview. The later study varies view count, output resolution and individual design choices.

Sections 2.1–2.3 and 3 · Main paper

Figures, captions and visual explanation

The diagram separates frozen VGGT features, latent compression, pointwise decoding and guidance. The resolution strip holds the latent fixed while increasing decoded points, making a design property visible rather than merely asserting flexibility.

Figure 2, PDF p. 3; Figure 5, p. 8 · Main paper

Tables, comparisons and evidence

Generated reference surfaces and native ground-truth surfaces have separate comparisons. Raw pointmaps are visually distinguished from globally fused surfaces, and the input-view budget is stated; these are essential conditions for interpreting ranks.

Tables 1–2; Table 2, PDF p. 8 · Main paper

Appendix and reproducibility

Related-work expansion, architecture settings, training surface preparation and guidance details are kept as named lookup routes. The main figure remains sufficient to locate the changed computation without reading every training setting.

Appendices A–D and evaluation details · Main paper

Clarity and writing conventions

Numbered ingredients and consistent encoder/decoder/guidance terminology create a short verbal map. Copy that consistency, not the assumption that the same three-part outline suits every architecture paper.

Figure 2 caption; Section 2 opening · Main paper

What to reuse—and what not to infer

Transferable lesson: Make an architectural invariant visible: identify what stays fixed, what scales, and which inference-time operations implement the intended property.

Boundary: The public title adds the name Surflo to the official program title; the author list matches. Its current arXiv rendering and results are not verified identical to the accepted copy.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-02

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-focused scan across framing, method organization, captions, result presentation and conclusion; title and author overlap checked against the official program. Selected physical pages inspected as rendered images.

Appendix / supplement

Attached appendix headings, reproduction routes and representative supporting prose/captions scanned. Long tables, every proof and every appendix page were not individually validated.

Visually inspected PDF pages

3, 8

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.

• Public author manuscript arXiv:2606.13644v1; not verified identical to the accepted OpenReview revision · 22 pages