Writing study: Importance-Based Token Merging for Efficient Image and Video Generation

ICCV overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICCV 2025 oral writing corpus

Haoyu Wu, Jingyi Xu, Hieu Le, Dimitris Samaras · Official ICCV 2025 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

13 physical pages

Before references / appendix

≈ 8.09 page-equivalents

Main figures / tables

8 / 9

Attached appendix figures / tables

0 / 0

Attached appendix span

No boundary detected; not assumed zero

Before references / appendix: ≈ 62.23%References / intervening material: ≈ 37.77%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	≈ 1.07 pp.	8.23%
2. Related Work	2	≈ 0.83 pp.	6.38%
3. Method	3	≈ 1.6 pp.	12.31%
4. Experiments	4	≈ 4.03 pp.	31%
5. Conclusions	8	≈ 0.21 pp.	1.62%
6. Acknowledgments	9	≈ 0.05 pp.	0.38%
References	9	≈ 4.91 pp.	37.77%

All detected figure and table locations

Main text

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

Attached appendix

No recognized caption labels.

Argument and organization

1. 1 Introduction; 2 Related Work

   Frames random destination-token choice as the quality weakness of otherwise useful token merging.

2. 3 Method

   Defines a CFG importance proxy, diagnoses naive top-k selection and introduces an importance pool with diversity-preserving matching.

3. 4 Experiments; 5 Conclusions

   Separates image, multiview and video generation, then reports resource costs and selection alternatives.

Problem, claim and contribution

The intervention reallocates inference computation among tokens rather than retraining the generative model.

Sections 1 and 3, physical pp.1–4 · Main paper

Argument and section order

A failed obvious solution, top-k destinations, is explained before the proposed pool, so the extra selection step has a concrete purpose.

Section 3 and Figure 4, pp.3–4 · Main paper

Figures, captions and visual explanation

A numbered-token example carries identity and importance shading through pooling, source/destination matching and the surviving independent token.

Figure 2, p.3 · Main paper

Tables, comparisons and evidence

Task-specific quality tables stay separate from the resource table; the latter distinguishes memory-efficient attention in parentheses, with A5000/FP16 and latent-space-only timing stated in the setup.

Tables 1–6, p.7; Section 4.1, pp.5–6 · Main paper

Appendix and reproducibility

Image/video prompts and multiview input images are delegated to supplementary material, so the static main examples alone do not provide fully reproducible sample conditions.

Figures 5–8 captions, pp.5–6,8 · Main paper

Clarity and writing conventions

The discussion admits that temporal token merging reduces generated dynamics, a failure not captured by simply emphasizing the total VBench score.

Section 4.2, p.7 · Main paper

What to reuse—and what not to infer

Transferable lesson: Use a small token identity trace to explain selection logic, and report efficiency under both optimized and unoptimized attention settings.

Boundary: CFG is an importance proxy, not an explanation-faithfulness guarantee. The free acceleration phrasing excludes setup-specific overhead and latent decoding; quality improvements are not uniform across every metric, such as multiview SSIM at high merging ratios.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-07

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan across the main manuscript, physical PDF pp.1–8, covering framing, section progression, method explanations, captions, comparison protocols and concluding scope; reference boundary checked. Selected full-page rasters were inspected. Not exhaustive line-by-line reading, proof verification or experimental reproduction.

Appendix / supplement

Main-paper appendix/supplement references were checked. A separate supplementary source is linked in the inventory but was not retrieved or inspected; its contents are not included in the reading or visual-page counts.

Visually inspected PDF pages

3, 7

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 CVF open-access ICCV 2025 accepted-paper PDF, retrieved 7 October 2026; CVF states identical to accepted version except watermark; not assumed identical to initial submission · 13 pages