Writing study: PR-MaGIC: Prompt Refinement Via Mask Decoder Gradient Flow For In-Context Segmentation

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Minjae Lee, Sungwoo Hur, Soojin Hwang, Won Hwa Kim · 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: Training, objectives & optimization (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

≈ 8.08 page-equivalents

Main figures / tables

7 / 2

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 80.8%References / intervening material: ≈ 19.2%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	≈ 1.48 pp.	14.8%
2. Related Works	2	≈ 0.52 pp.	5.2%
3. Method	3	≈ 2.91 pp.	29.1%
4. Experiments	6	≈ 2.55 pp.	25.5%
5. Conclusion	8	≈ 0.26 pp.	2.6%
References	9	≈ 1.92 pp.	19.2%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Figure 3 · p. 4
• Figure 4 · p. 5
• Figure 5 · p. 7
• Table 1 · p. 7
• Table 2 · p. 8
• Figure 6 · p. 8
• Figure 7 · 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 11 physical pages; 9 detected figures and 4 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 10.87 page-equivalents, 9 figure labels and 4 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.03 pp.
A. Proof of Proposition 1	1	≈ 0.93 pp.
B. MIoU Progression	2	≈ 1 pp.
C. One-shot segmentation results across random seeds	3	≈ 1 pp.
D. Qualitative results for top-1 selection	4	≈ 1.95 pp.
E. Additional Qualitative Results	6	≈ 5.05 pp.
F. Sensitivity analysis of γ and η across object categories.	11	≈ 0.91 pp.

• Table 1 · p. 2
• Table 2 · p. 3
• Table 3 · p. 3
• Figure 1 · p. 4
• Figure 2 · p. 5
• Figure 3 · p. 6
• Figure 4 · p. 6
• Figure 5 · p. 7
• Figure 6 · p. 7
• Figure 7 · p. 8
• Figure 8 · p. 9
• Figure 9 · p. 10
• Table 4 · p. 11

Argument and organization

1. Method §§3.1–3.4

   Present the mathematical flow, practical update, selection rule and qualified convergence motivation.

2. Experiments §§4.1–4.4

   Separate achievable top-1 results from oracle selection, then examine sensitivity and convergence failures.

Problem, claim and contribution

The motivating examples identify misplaced prompts as a failure remaining after automatic in-context prompting, so the intervention is refinement rather than another full segmentor.

PDF pp. 1–2, Figures 1–2 · Main paper · Additional source 2

Argument and section order

The formal flow is followed by algorithmic refinement and candidate selection, then Proposition 1 states a local proximity assumption rather than promising global convergence.

§§3.1–3.4, PDF pp. 3–6; Algorithm 1, PDF p. 5 · Main paper · Additional source 2

Figures, captions and visual explanation

The overview explicitly notes the omitted SAM prompt encoder and marks the proposed components; refinement figures show intermediate steps, while failure examples retain cases where later steps deteriorate.

Figure 3, PDF p. 4; Figures 4 and 7, PDF pp. 5 and 8 · Main paper · Additional source 2

Tables, comparisons and evidence

The main table labels baseline, oracle and top-1 separately, and the text defines oracle as the best mask among five steps. Iteration variability is quantified rather than assuming the last step is always best.

PDF p. 6, evaluation; Tables 1–2, PDF pp. 7–8 · Main paper · Additional source 2

Appendix and reproducibility

The supplement pairs the proof with per-step gains, random-seed variance and full refinement trajectories showing selected steps, making both theoretical and empirical qualifications traceable.

Supplement §§A–F, PDF pp. 1–11, Tables 1–4 and Figures 1–9 · Main paper · Additional source 2

Clarity and writing conventions

The theory-to-practice transition explicitly says the proximity assumption may fail; this motivates selection and sensitivity studies rather than silently extending the proposition to every example.

PDF p. 6, Proposition 1 and top-1 motivation · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: Keep oracle upper bounds visually distinct from deployable selection and place a theorem's assumptions beside the empirical failures that motivate safeguards.

Boundary: Training-free refinement still performs iterative computation; local continuous-flow convergence does not certify globally optimal masks or monotonic improvement of the discrete implementation.

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. 1–11, 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 §§A–F, PDF pp. 1–11, Tables 1–4 and Figures 1–9. 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 · 10 pages
• Official CVPR 2026 CVF supplementary PDF; retrieved 2026-09-26 · 11 pages