Writing study: Representation Fréchet Loss for Visual Generation

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Jiawei Yang, Zhengyang Geng, Xuan Ju, Yonglong Tian, Yue Wang · 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: 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

30 physical pages

Before references / appendix

≈ 10.23 page-equivalents

Main figures / tables

7 / 4

Attached appendix figures / tables

8 / 11

Attached appendix span

≈ 16.91 page-equivalents

Before references / appendix: ≈ 34.1%References / intervening material: ≈ 9.53%Attached appendix: ≈ 56.37%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	≈ 0.99 pp.	3.3%
2 Related Work	2	≈ 0.86 pp.	2.87%
3 Method	3	≈ 2.71 pp.	9.03%
4 Experiments	6	≈ 4.28 pp.	14.27%
5 Conclusion	10	≈ 0.45 pp.	1.5%
Acknowledgments	11	≈ 0.14 pp.	0.47%
References	11	≈ 2.86 pp.	9.53%
Appendix for	14	≈ 0.24 pp.	0.8%
A Additional Design Attempts and Observations	14	≈ 1.16 pp.	3.87%
B Implementation and Evaluation Details	15	≈ 2.25 pp.	7.5%
C Human Preference Study	17	≈ 1.02 pp.	3.4%
D Limitations and Broader Impact	18	≈ 0.43 pp.	1.43%
E Additional Qualitative Samples	19	≈ 4.9 pp.	16.33%
F Detailed Results	24	≈ 5.6 pp.	18.67%
G Text-to-Image Prompts	29	≈ 1.31 pp.	4.37%

All detected figure and table locations

Main text

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

Attached appendix

• Figure A.1 · p. 15
• Table B.1 · p. 15
• Table B.2 · p. 16
• Table B.3 · p. 17
• Figure C.1 · p. 18
• Figure E.1 · p. 20
• Figure E.2 · p. 21
• Figure E.3 · p. 22
• Figure E.4 · p. 23
• Table F.1 · p. 24
• Table F.2 · p. 24
• Table F.3 · p. 25
• Table F.4 · p. 25
• Table F.5 · p. 25
• Table F.6 · p. 26
• Table F.7 · p. 27
• Table F.8 · p. 28
• Figure G.1 · p. 29
• Figure G.2 · p. 29

Argument and organization

1. Problem and contribution

   Turns an evaluation statistic into a training objective by isolating the population-estimation versus gradient-memory bottleneck.

2. Argument sequence

   Defines FD and its estimators before population-size, representation, one-step conversion and system-level experiments.

3. Evidence and verification

   Population and feature-space ablations are separated from model-scale comparisons; a human study tests whether metric gains agree with preference.

Problem, claim and contribution

Turns an evaluation statistic into a training objective by isolating the population-estimation versus gradient-memory bottleneck.

Introduction · Main paper

Argument and section order

Defines FD and its estimators before population-size, representation, one-step conversion and system-level experiments.

Sections 3–4 · Main paper

Figures, captions and visual explanation

A compact queue/batch diagram explains the computational split. Paired images carry both FID and multi-representation scores, making metric disagreement visible.

Figure 2, p. 2; Figures 4–5, p. 8 · Main paper

Tables, comparisons and evidence

Population and feature-space ablations are separated from model-scale comparisons; a human study tests whether metric gains agree with preference.

Tables 1–4; Figure 6 · Main paper

Appendix and reproducibility

Unsuccessful design attempts, exact feature identifiers, unified evaluation and human-study protocol accompany uncurated sample pages.

Appendices A–F · Main paper

Clarity and writing conventions

The text explicitly rejects the inference that a low Inception FID means generation is solved.

Figure 3; Section 4.4 · Main paper

What to reuse—and what not to infer

Transferable lesson: When optimizing an evaluation metric, show held-out representations and human evidence, not just improvement on that metric.

Boundary: The alternative aggregate metric is still representation-dependent; author-version sample quality judgments are not universal perceptual guarantees.

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-oriented distributed manuscript scan of Sections 1–5, pp. 1–10; references pp. 11–13: abstract/framing, section sequence, substantive text windows, captions, evidence organization and conclusion; not exhaustive technical verification.

Appendix / supplement

Writing-oriented scan of Appendices A–G, pp. 14–30; design attempts, evaluation, human study, limitations and samples; code and experiments were not executed.

Visually inspected PDF pages

2, 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 arXiv manuscript 2604.28190v1; not verified identical to the accepted or camera-ready OpenReview revision · 30 pages