Writing study: CLEAR: Context-Aware Learning with End-to-End Mask-Free Inference for Adaptive Video Subtitle Removal

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Qingdong He, Chaoyi Wang, Peng Tang, Yifan Yang, Xiaobin Hu · Official ICML 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: Domain-grounded methods & physical 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

14 physical pages

Before references / appendix

≈ 9.11 page-equivalents

Main figures / tables

6 / 3

Attached appendix figures / tables

4 / 0

Attached appendix span

≈ 2.91 page-equivalents

Before references / appendix: ≈ 65.07%References / intervening material: ≈ 14.14%Attached appendix: ≈ 20.79%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: The physical p.12 PNG shows two body-size bold appendix headings omitted by the size filter: A at y=89.521806/792 and B at y=378.069806/792. Retain the existing Appendix start, and restore A/B in the single-column page coordinates. Newly detected unnumbered impact/singular-acknowledgment statements are retained from the size-aware source extraction; authored heading corrections are unchanged.

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	≈ 1.14 pp.	8.14%
2. Related Work	3	≈ 0.52 pp.	3.71%
3. Method	3	≈ 2.86 pp.	20.43%
4. Experiments	6	≈ 3.24 pp.	23.14%
5. Conclusion	9	≈ 0.2 pp.	1.43%
Impact Statement	10	≈ 0.07 pp.	0.5%
References	10	≈ 1.98 pp.	14.14%
Appendix	12	≈ 0.03 pp.	0.21%
A. Two-Stage Training Pipeline	12	≈ 0.36 pp.	2.57%
B. More Qualitative Results	12	≈ 2.52 pp.	18%

All detected figure and table locations

Main text

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

Attached appendix

• Figure A1 · p. 13
• Figure A2 · p. 13
• Figure A3 · p. 14
• Figure A4 · p. 14

Argument and organization

1. Problem and contribution

   Enumerates three limitations, three required capabilities and three corresponding innovations, giving the application constraints a traceable method mapping.

2. Argument sequence

   Prior learning and adaptive weighting are separated from deployment, with an explicit inference subsection before evaluation.

3. Evidence and verification

   The main numerical table states rank, sampling steps and configuration; it preserves slower elapsed inference than two baselines despite stronger quality metrics.

Problem, claim and contribution

Enumerates three limitations, three required capabilities and three corresponding innovations, giving the application constraints a traceable method mapping.

Introduction, p. 2 · Main paper

Argument and section order

Prior learning and adaptive weighting are separated from deployment, with an explicit inference subsection before evaluation.

Sections 3.2-3.4, pp. 3-6 · Main paper

Figures, captions and visual explanation

The overview has distinct training and inference lanes; the results page groups metrics by reconstruction, perception, temporal behavior, flow and time beside matched qualitative crops.

Figure 2, p. 2; Table 1 and Figure 5, p. 8; both PNGs inspected · Main paper

Tables, comparisons and evidence

The main numerical table states rank, sampling steps and configuration; it preserves slower elapsed inference than two baselines despite stronger quality metrics.

Table 1, p. 8 · Main paper

Appendix and reproducibility

The appendix repeats the two-stage contract and captions the displayed mask overlays as training-time guidance, not required inference inputs.

Appendix A, p. 12; Figures A3-A4, p. 14 · Main paper

Clarity and writing conventions

Named limitations/capabilities make the introduction navigable, but claims of complete residue-free removal and language-agnostic behavior outrun the selected qualitative examples.

Introduction, p. 2; qualitative/cross-lingual discussion, p. 8 · Main paper

What to reuse—and what not to infer

Transferable lesson: Separate training supervision, internal guidance and deployment inputs in both prose and the pipeline diagram.

Boundary: No manual mask annotation does not mean unpaired or supervision-free training; static example frames alone cannot verify temporal consistency or universal cross-lingual removal.

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-focused distributed manuscript scan, not exhaustive technical/proof validation. Sections 1-5, pp. 1-9: subtitle-specific constraints, two-stage training, mask-free inference, comparisons and ablations.

Appendix / supplement

Writing-focused appendix scan; code, data and external assets not executed. Appendices A-B, pp. 12-14: training/inference equations and extra qualitative/cross-lingual examples.

Visually inspected PDF pages

2, 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.

• Published ICML 2026 proceedings PDF, PMLR volume 306; retrieved 2026-10-07. Public OpenReview decision notes were not independently reverified. · 14 pages