Skip to content
Computational Pathology Reading Roadmap

Writing study: FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models

← ICCV 2025 oral writing corpus

Vladimir Kulikov, Matan Kleiner, Inbar Huberman-Spiegelglas, Tomer Michaeli · Official ICCV 2025 oral record

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
≈ 8.09 page-equivalents
Main figures / tables
8 / 0
Attached appendix figures / tables
0 / 0
Attached appendix span
No boundary detected; not assumed zero
Before references / appendix: ≈ 80.9%References / intervening material: ≈ 19.1%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 headingStarts on PDF pageEstimated spanShare of entire PDF
1. Introduction2≈ 0.35 pp.3.5%
2. Related work2≈ 0.36 pp.3.6%
3. Preliminaries2≈ 1.06 pp.10.6%
4. Reinterpretation of editing by inversion3≈ 0.81 pp.8.1%
5. FlowEdit4≈ 2.69 pp.26.9%
6. Experiments7≈ 1.51 pp.15.1%
7. Conclusion and limitations8≈ 0.22 pp.2.2%
Acknowledgements9≈ 0.04 pp.0.4%
References9≈ 1.91 pp.19.1%
All detected figure and table locations

Main text

Attached appendix

    No recognized caption labels.

    Argument and organization

    1. 1 Introduction; 2 Related Work; 3 Preliminaries

      Distinguishes inversion, optimization and intervention in model internals.

    2. 4 Reinterpretation of Editing by Inversion; 5 FlowEdit

      Rewrites the baseline as a direct path before introducing random-pairing velocity averaging and practical sampling.

    3. 6 Experiments; 7 Conclusion and Limitations

      Uses an exact-inversion synthetic control and per-backbone edit tradeoffs, then names large-edit limitations.

    Problem, claim and contribution

    The motivation notes that known exact noise does not itself preserve image structure, separating inversion error from an undesirable source/target pairing.

    Sections 1 and 5.3, physical pp.2,6 · Main paper

    Argument and section order

    Reinterpreting the existing method first makes the new ODE a single conceptual change; an explicit heuristic caveat precedes the implementation algorithm.

    Sections 4–5 and Algorithm 1, pp.3–5 · Main paper

    Figures, captions and visual explanation

    A three-panel diagram contrasts inversion through noise, its direct-path reinterpretation and FlowEdit, reusing colored velocities and intermediate images.

    Figure 2, p.3 · Main paper

    Tables, comparisons and evidence

    Separate SD3 and FLUX tradeoff plots connect hyperparameter settings rather than reporting one unqualified best score; the synthetic cat-to-dog control uses known generation noise.

    Figure 7, p.8; Section 5.3, pp.6–7 · Main paper

    Appendix and reproducibility

    Hyperparameters, style-edit changes, optional source prompts, optimization analysis and synthetic controls are routed to named appendices B,D,I,L,M; these were not inspected.

    Sections 5.2–6, pp.5–8 · Main paper

    Clarity and writing conventions

    The method explicitly says its heuristic neither guarantees an exact target distribution nor necessarily gives optimal transport, despite the transport-based motivation.

    Section 5, p.4 · Main paper

    What to reuse—and what not to infer

    Transferable lesson: Re-express a familiar baseline before introducing a change, and use an exact-control experiment to isolate the mechanism from implementation error.

    Boundary: Lower measured transport cost is not a proof of optimal transport or universal semantic preservation. Fidelity can oppose substantial edits; the quantitative cat/dog distribution is model-generated, and architecture-agnostic implementation has been demonstrated on two specified flow backbones, not every generator.

    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.