Writing study: NuWa: Deriving Lightweight Class-Specific Vision Transformers for Edge Devices

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Ziteng Wei, Qiang He, Bing Li, Feifei Chen, Hai Jin, Yun Yang · 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: Efficiency, inference & systems (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

≈ 8.09 page-equivalents

Main figures / tables

12 / 4

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 73.55%References / intervening material: ≈ 26.45%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	≈ 1.38 pp.	12.55%
2. Related Works	3	≈ 0.59 pp.	5.36%
3. Methodology	3	≈ 2.19 pp.	19.91%
4. Experiments and Analysis	5	≈ 2.9 pp.	26.36%
5. Conclusion	8	≈ 0.22 pp.	2%
Acknowledgments	9	≈ 0.04 pp.	0.36%
References	9	≈ 2.91 pp.	26.45%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Figure 3 · p. 2
• Figure 4 · p. 3
• Figure 5 · p. 4
• Table 1 · p. 6
• Figure 6 · p. 6
• Table 2 · p. 7
• Figure 7 · p. 7
• Figure 8 · p. 7
• Figure 9 · p. 7
• Table 3 · p. 8
• Table 4 · p. 8
• Figure 10 · p. 8
• Figure 11 · p. 8
• Figure 12 · 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 6 physical pages; 3 detected figures and 5 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 5.93 page-equivalents, 3 figure labels and 5 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.09 pp.
6. Calculation of FLOPs	1	≈ 0.14 pp.
7. Limitations of Compression Methods	1	≈ 0.55 pp.
8. Importance Metrics for Pruning	1	≈ 0.48 pp.
9. Ablation and Design Justification	2	≈ 0.91 pp.
10. Proof	3	≈ 1.66 pp.
11. Implementation Details	4	≈ 0.26 pp.
12. Baselines	5	≈ 0.59 pp.
13. Performance under Complex Settings	5	≈ 0.3 pp.
14. Pseudocode	6	≈ 0.25 pp.
15. Limitations and Future Works	6	≈ 0.71 pp.

• Table 5 · p. 1
• Figure 13 · p. 1
• Table 6 · p. 2
• Figure 14 · p. 2
• Table 7 · p. 2
• Table 8 · p. 3
• Table 9 · p. 4
• Figure 15 · p. 5

Argument and organization

1. Problem and contribution

   Uses the counterintuitive observation that pruning can improve target-class accuracy to question all-class compression objectives.

2. Argument sequence

   Separates self-knowledge purification from the closed-form fast-pruning stage, then evaluates derivation and deployment independently.

3. Evidence and verification

   Table 1 groups target-class subsets and pruning rates and marks fine-tuned variants; Table 3 measures latency, throughput and memory on both edge and desktop hardware.

Problem, claim and contribution

Uses the counterintuitive observation that pruning can improve target-class accuracy to question all-class compression objectives.

Figure 1 and Introduction, pp. 1–2 · Main paper · Additional source 2

Argument and section order

Separates self-knowledge purification from the closed-form fast-pruning stage, then evaluates derivation and deployment independently.

Sections 3.2–4.2, pp. 4–8 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 contrasts three derivation settings in parallel panels; the adjacent cost bars split pruning from retraining time.

Figures 2–3, p. 2 · Main paper · Additional source 2

Tables, comparisons and evidence

Table 1 groups target-class subsets and pruning rates and marks fine-tuned variants; Table 3 measures latency, throughput and memory on both edge and desktop hardware.

Tables 1 and 3, pp. 6 and 8 · Main paper · Additional source 2

Appendix and reproducibility

The supplement includes closed-form proof, actual pseudocode and co-occurrence/class-mismatch analysis rather than only additional headline compression ratios.

Supplement Sections 10–15, pp. 3–6 · Main paper · Additional source 2

Clarity and writing conventions

Training-free and training-dependent baselines receive separate result paragraphs, making the comparison contract explicit.

Section 4.1, p. 6 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: Differentiate one-time model-construction cost from repeated deployment cost, and expose fine-tuning in variant names.

Boundary: Target-class specialization changes the task; higher subset accuracy is not evidence of improvement on the original full-class problem.

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

Writing-oriented scan throughout Sections 1–5, PDF pp. 1–8; references pp. 9–11: abstract/framing, section sequence, figure/table captions, evaluation organization and conclusions; not an exhaustive technical verification.

Appendix / supplement

Writing-oriented supplement scan: Official ZIP-contained supplement Sections 6–15, pp. 1–6: FLOPs, design ablations, proof, implementation, complex settings, pseudocode and limits. Video/code assets were not executed or assessed.

Visually inspected PDF pages

2, 6

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 CVF open-access CVPR 2026 accepted-paper PDF · 11 pages
• Official CVF supplementary ZIP member PDF; see URL fragment for member · 6 pages