Writing study: Understanding Task Transfer in Vision-Language Models

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Bhuvan Sachdeva, Karan Uppal, Abhinav Java, Vineeth Balasubramanian · 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: Empirical analysis, diagnosis & auditing (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

10 / 2

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	1	≈ 1.06 pp.	10.6%
2. Related Work	2	≈ 0.78 pp.	7.8%
3. Problem Formulation and Metric	3	≈ 1.27 pp.	12.7%
4. Results and Analysis	4	≈ 1.32 pp.	13.2%
5. Additional Results	6	≈ 1.69 pp.	16.9%
6. Discussion	7	≈ 0.95 pp.	9.5%
7. Conclusion	8	≈ 0.22 pp.	2.2%
References	9	≈ 1.95 pp.	19.5%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Table 1 · p. 4
• Table 2 · p. 4
• Figure 3 · p. 5
• Figure 4 · p. 5
• Figure 5 · p. 6
• Figure 6 · p. 6
• Figure 7 · p. 7
• Figure 8 · p. 7
• Figure 9 · p. 8
• Figure 10 · 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 17 physical pages; 25 detected figures and 2 tables. These may include overlapping main-text material.

Detected heading	PDF page	Estimated span
References	17	≈ 0.91 pp.

• Figure A.2 · p. 2
• Figure A.3 · p. 2
• Figure A.4 · p. 3
• Figure A.5 · p. 3
• Figure A.6 · p. 4
• Figure A.7 · p. 4
• Figure A.8 · p. 5
• Figure A.9 · p. 5
• Figure A.10 · p. 6
• Table A.1 · p. 6
• Figure A.11 · p. 7
• Figure A.12 · p. 8
• Figure A.13 · p. 8
• Figure A.14 · p. 9
• Figure A.15 · p. 9
• Figure A.16 · p. 10
• Figure A.17 · p. 10
• Table A.2 · p. 11
• Figure A.18 · p. 12
• Figure A.19 · p. 13
• Figure A.20 · p. 13
• Figure A.21 · p. 14
• Figure A.22 · p. 14
• Figure A.23 · p. 15
• Figure A.24 · p. 15
• Figure A.25 · p. 16
• Figure A.26 · p. 16

Argument and organization

1. Define the measurement

   Introduce transferability/malleability and normalize changes by remaining performance headroom.

2. Analyze then apply

   Study task groups/model scale, interpret transfer graphs and test PGF-informed data mixtures.

Problem, claim and contribution

Unpredictable cross-task effects motivate measuring the full transfer map rather than adding another task-specific fine-tuning trick.

Abstract and Introduction · Main paper · Additional source 2

Argument and section order

The metric definition and toy example precede empirical patterns, keeping a novel statistic interpretable before its heatmaps are discussed.

Section 3 and Table 1, PDF pp. 3–4 · Main paper · Additional source 2

Figures, captions and visual explanation

A small numerical example distinguishes raw gain from gap-normalized improvement; later heatmaps and a grouped bar plot connect transfer patterns to data-selection outcomes.

Table 1, PDF p. 4; Figures 7–8, p. 7 · Main paper · Additional source 2

Tables, comparisons and evidence

Task categories are explicit, and the data-mixture experiment retains baseline, random, direct-supervision and PGF-informed conditions instead of comparing only the preferred mixture.

Table 2 and Figure 8 · Main paper · Additional source 2

Appendix and reproducibility

The supplement exposes raw accuracy, repeated-run variability, algorithmic metric definition and alternative ceilings, important for interpreting a normalized measure.

Supplement Sections A.1–A.10 · Main paper · Additional source 2

Clarity and writing conventions

Short Key Takeaway boxes summarize each analysis, while the limitations state that multiple-choice tasks restrict the observed transfer behaviors.

PDF pp. 7–8 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For empirical-analysis papers, teach a new metric with a tiny worked table, then link broad pattern maps to a controlled practical consequence.

Boundary: PGF depends on the chosen ceiling and can be sensitive near that ceiling; multiple-choice transfer results and preliminary mixture tests do not establish universal open-ended task relationships.

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 main PDF pp. 1–8: framing, section structure, method exposition, figure/table captions, evaluation organization and conclusions; not an exhaustive technical proof audit or reproduction.

Appendix / supplement

Separate supplement pp. 1–17 scanned using its contents, section prose and labeled matrices: PGF pseudocode, uncertainty/accuracy heatmaps, task groups, cliques, setup, training steps, LoRA analysis, model transfer and ceiling variants.

Visually inspected PDF pages

4, 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 CVPR 2026 CVF proceedings PDF; retrieved 2026-09-26 · 10 pages
• Official CVPR 2026 separate supplementary PDF; retrieved 2026-09-26 · 17 pages