Writing study: Understanding Co-speech Gestures in-the-wild

ICCV overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICCV 2025 oral writing corpus

Sindhu B Hegde, K R Prajwal, Taein Kwon, Andrew Zisserman · 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: Datasets, benchmarks & measurement (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

11 physical pages

Before references / appendix

≈ 8.13 page-equivalents

Main figures / tables

4 / 7

Attached appendix figures / tables

0 / 0

Attached appendix span

No boundary detected; not assumed zero

Before references / appendix: ≈ 73.91%References / intervening material: ≈ 26.09%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 pp.	9.09%
2. Related Work	2	≈ 1.03 pp.	9.36%
3. Method	3	≈ 1.91 pp.	17.36%
4. Downstream Tasks and Evaluation	5	≈ 1.14 pp.	10.36%
5. Results	6	≈ 1.35 pp.	12.27%
6. Insights and Ablations	7	≈ 1.11 pp.	10.09%
7. Conclusion	8	≈ 0.27 pp.	2.45%
References	9	≈ 2.87 pp.	26.09%

All detected figure and table locations

Main text

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

Attached appendix

No recognized caption labels.

Argument and organization

1. 1 Introduction; 2 Related Work

   Explains sparse and ambiguous speech–gesture correspondence before introducing three understanding tasks.

2. 3 Method

   Separates modality encoders, word alignment, phrase contrast and local coupling, then documents training sources.

3. 4 Downstream Tasks and Evaluation; 5 Results; 6 Insights; 7 Conclusion

   Defines curated task protocols before scores, then isolates speech/text complementarity and objective/fusion choices.

Problem, claim and contribution

Explicitly states that some gestures have no semantic word counterpart and that phrase-level supervision does not identify gestured words.

Sections 1 and 3.7, physical pp.1–2,4 · Main paper

Argument and section order

New task definitions and benchmark curation precede results, preventing the method-specific embedding scores from substituting for a clear evaluation contract.

Sections 4.1–4.4, pp.5–6 · Main paper

Figures, captions and visual explanation

The architecture separates phrase pooling from word alignment; the later heatmap overlays speech boundaries and gesture frames to explain why alignment must extend beyond spoken-word intervals.

Figure 2, p.3; Figure 3, p.8 · Main paper

Tables, comparisons and evidence

The retrieval benchmark fixes a 500-clip gallery, while active-speaker tests vary the number of candidates. The discussion acknowledges GestSync wins the synchronization-sensitive task.

Section 4.4, p.6; Table 4 and Section 5.4, p.7 · Main paper

Appendix and reproducibility

Additional implementation and AVS-Spot examples/results are routed to supplementary material; only the main descriptions and pointers were inspected.

Sections 3.8 and 4.4, pp.4,6 · Main paper

Clarity and writing conventions

Modality analysis distinguishes semantic word information from acoustic stress, explaining task-dependent winners instead of forcing a universal multimodal advantage.

Section 6.1 and Table 5, pp.7–8 · Main paper

What to reuse—and what not to infer

Transferable lesson: For weakly aligned modalities, visualize annotation-time boundaries against the event being localized and report task-specific non-wins.

Boundary: Curated clips with visible gesture activity and clean speech/transcripts are not an unconstrained communication population. Candidate-set speaker matching is not full multi-person conversational diarization; plausible heatmaps do not prove language comprehension.

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, 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.

• 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 · 11 pages