Writing study: Generating metamers of human scene understanding

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Ritik Raina, Abe Leite, Alexandros Graikos, Seoyoung Ahn, Dimitris Samaras, Gregory Zelinsky · Official ICLR 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: Empirical analysis, diagnosis & auditing (primary) · Architecture & representation design.

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

26 physical pages

Before references / appendix

≈ 10.39 page-equivalents

Main figures / tables

5 / 0

Attached appendix figures / tables

11 / 0

Attached appendix span

≈ 10.89 page-equivalents

Before references / appendix: ≈ 39.96%References / intervening material: ≈ 18.15%Attached appendix: ≈ 41.88%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

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	≈ 1.16 pp.	4.46%
2 PRELIMINARIES	2	≈ 0.98 pp.	3.77%
3 MetamerGen: PERCEPTUALLY-INFORMED CONDITIONING	3	≈ 2.41 pp.	9.27%
4 BEHAVIORALLY-CONDITIONED SCENE METAMERS	6	≈ 1.29 pp.	4.96%
5 RESULTS: PROBING THE FEATURES OF SCENE METAMERS	7	≈ 3.05 pp.	11.73%
6 LIMITATIONS	10	≈ 0.17 pp.	0.65%
7 DISCUSSION	10	≈ 0.42 pp.	1.62%
ETHICS STATEMENT	11	≈ 0.15 pp.	0.58%
ACKNOWLEDGMENTS	11	≈ 0.14 pp.	0.54%
REFERENCES	11	≈ 4.71 pp.	18.12%
A APPENDIX	16	≈ 10.89 pp.	41.88%

All detected figure and table locations

Main text

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

Attached appendix

• Figure 6 · p. 16
• Figure 7 · p. 18
• Figure 8 · p. 18
• Figure 9 · p. 19
• Figure 10 · p. 20
• Figure 11 · p. 21
• Figure 12 · p. 22
• Figure 13 · p. 23
• Figure 14 · p. 24
• Figure 15 · p. 25
• Figure 16 · p. 26

Argument and organization

1. Introduction and Background

   Connect cognitive-science scene metamers to diffusion, conditioning adapters, and visual encoders.

2. MetamerGen and Behavioral Experiment

   Translate foveal/peripheral hypotheses into dual conditioning streams and an operational same/different judgment.

3. Results, Limitations, Discussion

   Separate model demonstrations, feature correlations, and ablations before interpreting human perception.

Problem, claim and contribution

The introduction makes an otherwise abstract internal-representation question experimentally tangible through visually different scenes receiving the same judgment.

Sections 1 and 4, pp. 1–2, 5–6 · Main paper

Argument and section order

A short technical background supports readers from cognitive science before the two visual streams are mapped to model components.

Sections 2–3, pp. 2–5 · Main paper

Figures, captions and visual explanation

The architecture uses a whole-pipeline view plus a magnified UNet block, with colors and frozen/trainable icons reused across levels.

Figure 1, p. 4 · Main paper

Tables, comparisons and evidence

Aligned feature-response panels retain correlation coefficients and uncertainty; their modest correlations should not be rewritten as causal explanations.

Figure 5 and Section 5.2, pp. 8–10 · Main paper

Appendix and reproducibility

The appendix gives gaze calibration, resampler configuration, generation ablations, and paired examples; it also explains why the ablation's metamer rate differs from the primary experiment.

Appendix A.1–A.10, pp. 16–26; Figure 13, p. 23 · Main paper

Clarity and writing conventions

The discussion preserves the nonsignificant human-versus-random fixation result, balancing the more positive feature-level interpretation.

Section 7, p. 10 · Main paper

What to reuse—and what not to infer

Transferable lesson: Tie each proposed conditioning stream to an experimental hypothesis, then display mechanism, behavioral protocol, and failure examples as distinct evidence layers.

Boundary: A same judgment after a five-second delay and 200 ms comparison is operational metamerism, not identical internal representations. Foveal tokens retain contextual information from full-image encoding. Appendix A.3's embedding width and A.4's reported variance changes need clarification; avoid copying their values uncritically.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–7; ethics and references pp. 11–15

Appendix / supplement

Appendix A.1–A.10, pp. 16–26

Visually inspected PDF pages

4, 9, 23

• ICLR 2026 official proceedings PDF, retrieved 2026-09-19 · 26 pages