Writing study: Steerable Vision Transformers
← ECCV 2026 oral writing corpus
Jona Ruthardt, Manu Gaur, Deva Ramanan, Makarand Tapaswi, Yuki Asano · Official ECCV 2026 oral record
Figure, table and section ledger
Editorial contribution type: Architecture & representation (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
- 17 physical pages
- Before references / appendix
- ≈ 14.32 page-equivalents
- Main figures / tables
- 10 / 4
- Attached appendix figures / tables
- 0 / 0
- Attached appendix span
- No boundary detected; not assumed zero
Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.
Source-bound correction: All top-level main headings and the reference boundary checked against source PDF text; source-bound positions use physical single-column y/page-height, not a two-column reading-order surrogate. Running page-number/author titles, affiliations, numbered contribution bullets, footnotes and bibliography continuations are excluded. This linked manuscript contains no appended supplementary section; separate supplement links were not scanned. Selected disputed heading locations also checked in rendered source pages.
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 | ≈ 1.99 pp. | 11.71% |
| 2 Related Work | 4 | ≈ 0.68 pp. | 4% |
| 3 SteerViT: Steering Vision Transformers with Text | 5 | ≈ 1.39 pp. | 8.18% |
| 4 Experiments | 6 | ≈ 8.01 pp. | 47.12% |
| 5 Conclusion | 14 | ≈ 0.52 pp. | 3.06% |
| Acknowledgements | 15 | ≈ 0.17 pp. | 1% |
| References | 15 | ≈ 2.68 pp. | 15.76% |
All detected figure and table locations
Main text
Attached appendix
No recognized caption labels.
Argument and organization
- 1 Introduction; 2 Related Work
Defines two desiderata—directable features and preserved visual representation quality—so a method that merely overwrites vision with text cannot count as success.
- 3 SteerViT: architecture, objective and data
A taxonomy precedes the model, and experiments are organized around distinct capabilities: conditional retrieval, localization, preservation, specificity and transfer.
- 4 Experiments by capability; 5 Conclusion
Steerability is tested alongside downstream feature quality, while a component table isolates early fusion, gating and foreground supervision rather than only reporting a large model.
Problem, claim and contribution
Defines two desiderata—directable features and preserved visual representation quality—so a method that merely overwrites vision with text cannot count as success.
Section 1, pp.2–3 · Main paper
Argument and section order
A taxonomy precedes the model, and experiments are organized around distinct capabilities: conditional retrieval, localization, preservation, specificity and transfer.
Figures 2–3, pp.2–3; Sections 4.1–4.7, pp.7–13 · Main paper
Figures, captions and visual explanation
The taxonomy aligns four visual-encoding families under common inputs and fusion points; the retrieval figure pairs embedding views with a labeled benchmark bar chart.
Figure 3 and Table 1, p.3; Figure 5, p.7 · Main paper
Tables, comparisons and evidence
Steerability is tested alongside downstream feature quality, while a component table isolates early fusion, gating and foreground supervision rather than only reporting a large model.
Sections 4.3 and 4.7, pp.9,13; Table 3, p.13 · Main paper
Appendix and reproducibility
Main-text pointers assign dataset mixtures, feature-extraction protocols, per-scene retrieval and extended layer analyses to the supplement; that separate PDF was not inspected.
Sections 3.3 and 4, pp.6–9,13–14 · Main paper
Clarity and writing conventions
The paper repeatedly returns to its two desiderata and uses capability-led headings rather than a single undifferentiated results section.
Section 1, p.3; Sections 4.1–4.6, pp.7–12 · Main paper
What to reuse—and what not to infer
Transferable lesson: When adding a capability to a foundation representation, measure both the new ability and retained original competence, and show where information enters.
Boundary: The event title Steerable Vision Transformers differs from the manuscript title Steerable Visual Representations. Synthetic retrieval construction and natural-image transfer are different evidence sources.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-02
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-oriented scan across the manuscript main argument, physical PDF pp.1–14: framing and section roadmaps, method/argument explanations, figure and table captions, comparison protocols, conclusions and limitations. Selected full-page rasters were inspected. This is not exhaustive line-by-line technical reading, proof verification or experimental reproduction.
- Appendix / supplement
- Main-manuscript pointers to appendix/supplement and the reference boundary were checked. Separate supplementary PDFs were not retrieved or inspected; their contents are not included in inspected-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.