Skip to content
Computational Pathology Reading Roadmap

Writing study: Automated Model Evaluation for Object Detection via Prediction Consistency and Reliability

← ICCV 2025 oral writing corpus

Seungju Yoo, Hyuk Kwon, Joong-Won Hwang, Kibok Lee · 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
≈ 8.12 page-equivalents
Main figures / tables
6 / 7
Attached appendix figures / tables
0 / 0
Attached appendix span
No boundary detected; not assumed zero
Before references / appendix: ≈ 81.2%References / intervening material: ≈ 18.8%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.

Source-bound correction: Raster checks on pp.4,7,8 restore Figure3, whose label is followed by a panel marker without a period, and the side-by-side Tables3 and5 lost in merged caption lines. Main has Figures1–7 and Tables1–7; positions use glyph coordinates except the visually located Figure3.

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. Introduction1≈ 1.3 pp.13%
2. Related Work2≈ 0.78 pp.7.8%
3. Prediction Consistency and Reliability3≈ 2.48 pp.24.8%
4. Corruption-Based Meta-Dataset5≈ 0.66 pp.6.6%
5. Experiments6≈ 2.29 pp.22.9%
6. Conclusion8≈ 0.21 pp.2.1%
Acknowledgements9≈ 0.07 pp.0.7%
References9≈ 1.88 pp.18.8%
All detected figure and table locations

Main text

Attached appendix

    No recognized caption labels.

    Argument and organization

    1. 1 Introduction; 2 Related Work

      Motivates detection-specific output geometry beyond classification-style covariate-shift proxies.

    2. 3 Prediction Consistency and Reliability; 4 Corruption-Based Meta-Dataset

      Defines complementary low/high-confidence box statistics and trains a dataset-level performance regressor.

    3. 5 Experiments; 6 Conclusion

      Crosses detectors and meta-datasets, then ablates severity, score components and target mAP definitions.

    Problem, claim and contribution

    Pre/post-NMS redundancy is repurposed as evaluation evidence; low-confidence consistent boxes and high-confidence reliable boxes are interpreted differently.

    Sections 1 and 3.1–3.2, physical pp.1–4 · Main paper

    Argument and section order

    Per-box statistics become per-image and then dataset-average scores before linear regression predicts mAP, exposing the aggregation hierarchy.

    Section 3.3, p.5 · Main paper

    Figures, captions and visual explanation

    The correlation panel pairs scatterplots with failure/success box examples and explicitly plots one minus the consistency score where a positive slope is shown.

    Figure 4, p.5 · Main paper

    Tables, comparisons and evidence

    Tables cross four detector/backbone combinations with augmentation and corruption meta-datasets; the protocol leaves a source dataset out for testing without transformations.

    Section 3.3, p.5; Tables 1–2, p.7 · Main paper

    Appendix and reproducibility

    Baseline tuning and further evaluation results are assigned to supplementary material, not inspected.

    Section 5.1, p.6 · Main paper

    Clarity and writing conventions

    The severity analysis describes increasing RMSE rather than implying that a lower error than the baseline means immunity to distribution shift.

    Figure 6 and Section 5.4, p.7 · Main paper

    What to reuse—and what not to infer

    Transferable lesson: Trace an evaluation proxy from individual predictions to the dataset-level target, and expose the calibration split and detector assumptions.

    Boundary: No ground-truth labels are needed for a new target dataset, but fitting and validating the performance regressor uses source meta-dataset mAP. Corruption coverage is not every real domain shift, and the pre-NMS mechanism does not automatically apply to NMS-free detectors or certify deployment safety.

    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
    5, 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.