Writing study: Confusion-Aware Spectral Regularizer for Long-Tailed Recognition

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Ziquan Zhu, Gaojie Jin, Hanruo Zhu, Si-Yuan Lu, Yunxiao Zhang, ZEYU FU, Ronghui Mu, Guoqiang Zhang, Zhao Sun, Yuhang Xia, Jiaxing Shang, Xiang Li, Lu Liu, Tianjin Huang · 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

11 physical pages

Before references / appendix

≈ 8.16 page-equivalents

Main figures / tables

3 / 7

Attached appendix figures / tables

0 / 0

Attached appendix span

No boundary detected; not assumed zero

Before references / appendix: ≈ 74.18%References / intervening material: ≈ 25.82%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	≈ 0.85 pp.	7.73%
2. Preliminaries	2	≈ 0.28 pp.	2.55%
3. Weighted Worst-Class Error	2	≈ 0.9 pp.	8.18%
4. CAR:Confusion-Aware Spectral Regularizer	3	≈ 0.58 pp.	5.27%
5. Experiments	4	≈ 3.09 pp.	28.09%
6. Ablation Studies	7	≈ 0.9 pp.	8.18%
7. Related Work	8	≈ 0.5 pp.	4.55%
8. Conclusion	8	≈ 0.25 pp.	2.27%
9. Acknowledgment	9	≈ 0.11 pp.	1%
References	9	≈ 2.84 pp.	25.82%

All detected figure and table locations

Main text

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

Attached appendix

No recognized caption labels.

Argument and organization

1. Metric before regularizer

   Define weighted worst-class error after preliminaries, then construct the confusion-aware regularizer.

2. Generalization and controlled mechanism tests

   Separate pretrained/from-scratch, backbones, imbalance, augmentation compatibility and component ablations.

Problem, claim and contribution

The opening emphasizes rare-class generalization rather than overall accuracy alone; the training/test worst-class gap becomes the motivating diagnostic.

Figure 1 and Sections 1–3 · Main paper

Argument and section order

Weighted worst-class error precedes the CAR construction, so the objective is introduced as a response to a defined failure measure.

Sections 2–4 · Main paper

Figures, captions and visual explanation

Figure 2 places matched confusion-matrix comparisons into two labeled groups for training from scratch and pretrained fine-tuning, retaining the same red intensity scale.

Figure 2, PDF p. 7 · Main paper

Tables, comparisons and evidence

Tables separate head/medium/tail/overall accuracy, worst-class train/test behavior, backbone transfer and imbalance factors rather than aggregating these into one headline.

Tables 1–6, PDF pp. 5–7 · Main paper

Appendix and reproducibility

Main-text references identify where stability analysis and setup details are intended to live, but unavailable appendix files must not be presented as inspected evidence.

Main PDF pp. 4 and 7, Appendix references 11–12 · Main paper

Clarity and writing conventions

Bold run-in headings name EMA estimation and each ablation, helping readers follow an equation-heavy objective into implementation decisions.

Section 4, PDF p. 4; Section 6, p. 7 · Main paper

What to reuse—and what not to infer

Transferable lesson: Introduce the failure metric before its surrogate loss and keep minority-group, worst-class and overall results distinct in the reporting layout.

Boundary: Confusion plots visualize selected settings; they do not replace repeated-run uncertainty or establish every class improves. The scan cannot verify the separately referenced appendix proofs.

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

No separate supplementary PDF appears in source metadata. Main text points to Appendices 11 and 12 for stability, datasets and additional results; those referenced materials were not available in this recorded download.

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 · 11 pages