Writing study: BEAST3D: animal behavioral analysis and neural encoding from multi-view video via Gaussian splatting

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Yanchen Wang, Lenny Aharon, Wangshu Zhu, Kyle Daruwalla, Linghua Zhang, Jiaru Zou, Selmaan N Chettih, Xun H Hou, Liam Paninski, Matthew Whiteway · 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: Domain-grounded methods & physical systems (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

31 physical pages

Before references / appendix

≈ 10.09 page-equivalents

Main figures / tables

5 / 0

Attached appendix figures / tables

4 / 8

Attached appendix span

≈ 15.91 page-equivalents

Before references / appendix: ≈ 32.55%References / intervening material: ≈ 16.13%Attached appendix: ≈ 51.32%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. Appendix span includes its explicit cover/contents/list-of-figures frontmatter; actual lettered subsection boundaries use their substantive starts, not TOC links.

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	≈ 1.46 pp.	4.71%
2 Related Work	3	≈ 1.35 pp.	4.35%
3 Method	4	≈ 1.6 pp.	5.16%
4 Results	6	≈ 3.9 pp.	12.58%
5 Conclusion	10	≈ 0.94 pp.	3.03%
References	11	≈ 5 pp.	16.13%
Supplementary Material	16	≈ 0.16 pp.	0.52%
A Datasets	16	≈ 2.45 pp.	7.9%
B Dataset construction	18	≈ 1.39 pp.	4.48%
C BEAST3D	20	≈ 2 pp.	6.45%
D Baselines	22	≈ 2.41 pp.	7.77%
E Novel view synthesis	24	≈ 2.59 pp.	8.35%
F Inference compute cost	27	≈ 0.34 pp.	1.1%
G Pose estimation	27	≈ 2.26 pp.	7.29%
H Neural encoding	29	≈ 1.12 pp.	3.61%
I Broader impacts	30	≈ 1.19 pp.	3.84%

All detected figure and table locations

Main text

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

Attached appendix

• Table 1 · p. 18
• Figure 6 · p. 23
• Table 2 · p. 24
• Table 3 · p. 25
• Table 4 · p. 25
• Table 5 · p. 26
• Table 6 · p. 26
• Figure 7 · p. 27
• Figure 8 · p. 28
• Table 7 · p. 29
• Figure 9 · p. 29
• Table 8 · p. 30

Argument and organization

1. Sparse-view laboratory problem

   Explains why generic reconstruction and annotated pose pipelines do not match the observed acquisition regime.

2. Representation and training

   Defines calibrated views and a held-out-view reconstruction task before the architecture and loss.

3. Three downstream evaluations

   Organizes evidence by novel-view synthesis, pose estimation and neural encoding, with protocol detail in the appendix.

Problem, claim and contribution

The paper motivates a specific sparse, calibrated multi-camera laboratory regime. This gives the reader a concrete reason why a generic 3D model and a pose-only representation are not automatically adequate baselines.

Introduction and related work · Main paper

Argument and section order

The method moves from problem setting to dataset construction, architecture and training. Results are divided by three downstream uses, making each evaluation answer a different representation-quality question.

Sections 3–4 · Main paper

Figures, captions and visual explanation

The left-to-right diagram shows raw views, frozen DINOv3 features, trainable multi-view attention, splats and rendered targets. Freeze/train symbols and repeated camera colors identify what is learned and what supervision is available.

Figure 2, PDF p. 4 · Main paper

Tables, comparisons and evidence

Neural encoding is shown at time-series, per-neuron and aggregate levels. Full reconstruction tables separate in-distribution and out-of-distribution evaluation and distinguish ground-truth versus rendered masks.

Figure 5, PDF p. 9; Tables 3–6 · Main paper

Appendix and reproducibility

Dataset and split construction, architecture, baselines, full novel-view results, pose learning-rate selection and neural-encoding protocols form task-specific lookup routes. The main text directs readers to the relevant route rather than one undifferentiated supplement.

Appendices A–H · Main paper

Clarity and writing conventions

Named paragraphs distinguish datasets, models, evaluation and results. The caption defines what one scatter point means and the error bars, preventing a population-level conclusion from being confused with one neuron or session.

Section 4.3; Figure 5 caption · Main paper

What to reuse—and what not to infer

Transferable lesson: For an applied representation paper, connect the acquisition constraints to the architecture and give each downstream evaluation its own question, protocol and unit of analysis.

Boundary: This author manuscript is a writing example, not proof that the reported results generalize to all species or laboratory setups. Scientific findings and every protocol were not independently reproduced.

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

4, 9

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:2606.02937v1; not verified identical to the accepted OpenReview revision · 31 pages