Writing study: SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching

← CVPR 2026 oral writing corpus

Yasaman Haghighi, Alex Alahi · Official CVPR 2026 oral record

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: 10 physical pages
Before references / appendix: ≈ 8.05 page-equivalents
Main figures / tables: 4 / 3
Attached appendix figures / tables: 0 / 0
Attached appendix span: Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 80.5%References / intervening material: ≈ 19.5%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	2	≈ 0.62 pp.	6.2%
2. It provides a sensitivity-based interpretation of why prior	2	≈ 0.02 pp.	0.2%
3. It adapts cache/reuse decisions per sample, unlike prior	2	≈ 0.02 pp.	0.2%
4. It requires no additional training and no model modifi-	2	≈ 0.02 pp.	0.2%
5. While our experiments focus on visual domain, the un-	2	≈ 5.87 pp.	58.7%
6. Discussions and Future Work	8	≈ 0.16 pp.	1.6%
7. Conclusion	8	≈ 0.16 pp.	1.6%
Acknowledgments	8	≈ 0.15 pp.	1.5%
References	9	≈ 1.95 pp.	19.5%

All detected figure and table locations

Main text

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 1 physical pages; 1 detected figures and 0 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 0.93 page-equivalents, 1 figure labels and 0 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.93 pp.

Figure 1 · p. 1

Argument and organization

Sensitivity diagnosis
Define the denoising dynamics before estimating sensitivity to latent and timestep changes.
Adaptive policy and measured scope
Translate the approximation into pseudocode, compare prior policies, then analyze calibration and cache lifetime.

Problem, claim and contribution

The contribution is framed as replacing heuristic reuse criteria with a local sensitivity estimate, explicitly naming both changing inputs: noisy latent and time.

Abstract and Section 4, PDF pp. 1 and 4–5 · Main paper · Additional source 2

Argument and section order

A relation-to-prior-methods subsection interprets existing policies through the new decomposition, helping readers understand both novelty and continuity.

Section 4.4, PDF p. 6 · Main paper · Additional source 2

Figures, captions and visual explanation

The calibration-size plot uses aligned model rows and separate latent/timestep sensitivity columns, making agreement between small and large calibration sets visually inspectable.

Figure 4, PDF p. 6; visual inspection · Main paper · Additional source 2

Tables, comparisons and evidence

The main comparison uses matched NFE, while separate tables vary maximum cache lifetime and tolerance. This exposes a regime where longer reuse stops improving compute but still worsens quality.

Tables 1–3, PDF p. 7; discussion p. 8 · Main paper · Additional source 2

Appendix and reproducibility

The one-page supplement targets a specific cross-model anomaly—higher required tolerance in CogVideoX and LTX—using adjacent-output MAE rather than repeating the main leaderboard.

Supplement Figure 1, PDF p. 1 · Main paper · Additional source 2

Clarity and writing conventions

The practical implementation states that sensitivities are precomputed per model on eight videos, qualifying what training-free entails operationally.

Section 4.5, PDF p. 6 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For an approximation-based accelerator, connect the mathematical surrogate to a concrete calibration protocol, then show where its local validity degrades.

Boundary: Training-free is not calibration-free, and matched NFE is not a direct measurement of wall-clock speed or absolute perceptual quality.

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. 1–10: 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. 1–1, 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 Figure 1, PDF p. 1. 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.