Writing study: SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Models

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Jiwoo Chung, Sangeek Hyun, MinKyu Lee, Byeongju Han, Geonho Cha, Dongyoon Wee, Youngjun Hong, Jae-Pil Heo · Official CVPR 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: Efficiency, inference & 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

12 physical pages

Before references / appendix

≈ 8.12 page-equivalents

Main figures / tables

10 / 4

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 67.67%References / intervening material: ≈ 32.33%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	≈ 1.51 pp.	12.58%
2. Related Work	2	≈ 0.52 pp.	4.33%
3. Preliminary	3	≈ 0.81 pp.	6.75%
4. Method: SeaCache	4	≈ 2.12 pp.	17.67%
5. Experiments	6	≈ 2.55 pp.	21.25%
6. Conclusion	8	≈ 0.21 pp.	1.75%
Acknowledgments	9	≈ 0.07 pp.	0.58%
References	9	≈ 3.88 pp.	32.33%

All detected figure and table locations

Main text

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

Attached appendix

No recognized caption labels.

Separate source files—excluded from canonical-PDF totals

A separate author version may repeat main text and references. Whole-file lengths and counts are not extra appendix material and are never added to the canonical PDF totals. Read the study's source scope for which sections received substantive reading versus layout inspection.

Additional source 1

Separate PDF (or ZIP containing it): whole-file layout of 7 physical pages; 2 detected figures and 6 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 6.88 page-equivalents, 2 figure labels and 6 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.22 pp.
1. Derivation of Optimal Linear Response	1	≈ 0.99 pp.
2. Runtime Overhead of SEA Filtering	2	≈ 0.22 pp.
3. Additional Evaluation	2	≈ 1.95 pp.
4. Limitation	4	≈ 2.55 pp.
References	7	≈ 0.95 pp.

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

Argument and organization

1. Oracle test before cache design

   Establish why raw feature distance is a poor reuse criterion through matched-refresh experiments.

2. Filter derivation to deployment evidence

   Define the spectral filter and refresh rule before testing image/video models, quality, cost and scheduling behavior.

Problem, claim and contribution

An oracle experiment compares raw and filtered cache criteria at matched refresh ratios, testing the premise before presenting an efficient input-side approximation.

Figure 2, PDF p. 2 · Main paper · Additional source 2

Argument and section order

The exposition separates the spectral filter from the caching policy, clearly stating that only the distance metric is replaced.

Sections 4.1–4.2 and Figure 3, PDF pp. 4–5 · Main paper · Additional source 2

Figures, captions and visual explanation

Page 6 places image and video results in parallel tables with latency, TFLOPs and reference-fidelity metrics; a separate rank table avoids mixing preference ranking into those physical units.

Tables 1–3, PDF p. 6; visual inspection · Main paper · Additional source 2

Tables, comparisons and evidence

The evaluation states that PSNR/LPIPS/SSIM use the corresponding full-compute output as reference, while CycleReward supplies a distinct perceptual comparison.

Evaluation protocol, PDF p. 6 · Main paper · Additional source 2

Appendix and reproducibility

Filter overhead is measured separately and the supplement makes linear-filter assumptions explicit, including stationarity and signal/noise independence.

Supplement Sections 1–2 and 4, PDF pp. 1–2 and 4 · Main paper · Additional source 2

Clarity and writing conventions

Refresh ratio is explicitly defined as full denoiser evaluations rather than vaguely called a cache percentage, clarifying what the budget controls.

Figure 2 caption, PDF p. 2 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For an approximate accelerator, compare at matched compute budgets, define the full-compute reference, and measure the added decision mechanism's overhead separately.

Boundary: Similarity to a no-cache sample measures preservation of a trajectory, not absolute generative quality. The optimal-linear-filter analysis is not an exact model of every nonlinear denoiser.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-26

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Manuscript-wide writing scan of PDF pp. 1–12: framing, argument sequence, section organization, figure/table captions, evaluation interpretation and limitations; reference sections identified. This is a writing-focused scan, not an exhaustive technical proof or reproduction audit. Main PDF pages 3 and 6 were rendered and visually inspected.

Appendix / supplement

Writing-focused scan of the separate supplementary PDF, pp. 1–7, including its organization, captions, protocol/evidence detail and limitation material where present; proofs and code listings were inspected for exposition, not independently verified or executed. Supplement Sections 1–2 and 4, PDF pp. 1–2 and 4. Linked video/audio demonstrations and non-PDF archive contents were not assessed.

Visually inspected PDF pages

3, 6

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 CVPR 2026 CVF open-access accepted manuscript; retrieved 2026-09-26 · 12 pages
• Official CVPR 2026 CVF supplementary PDF; retrieved 2026-09-26 · 7 pages