Writing study: Rethinking Dataset Distillation: Hard Truths about Soft Labels

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Priyam Dey, Aditya Sahdev, Sunny Bhati, Konda Reddy Mopuri, R. Venkatesh Babu · 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: 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

≈ 8.05 page-equivalents

Main figures / tables

3 / 3

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 80.5%References / intervening material: ≈ 19.5%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	≈ 1.63 pp.	16.3%
2. Preliminaries and Setup	3	≈ 0.33 pp.	3.3%
3. Label regimes of Data-Efficient Learning	3	≈ 3.21 pp.	32.1%
4. Distillation Correlation Score (DCS)	6	≈ 0.98 pp.	9.8%
5. Proposed Method	7	≈ 0.92 pp.	9.2%
6. Conclusion	8	≈ 0.14 pp.	1.4%
7. Future work and limitations	8	≈ 0.2 pp.	2%
References	9	≈ 1.95 pp.	19.5%

All detected figure and table locations

Main text

• Figure 1 · p. 2
• Table 1 · p. 4
• Figure 2 · p. 5
• Table 2 · p. 6
• Figure 3 · p. 7
• Table 3 · 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 17 physical pages; 10 detected figures and 11 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 16.94 page-equivalents, 10 figure labels and 11 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.02 pp.
1. Details on Methods and their loss objectives	1	≈ 2.62 pp.
2. Details on Training and Hyperparameters	3	≈ 0.35 pp.
3. Fitting scaling law equation for label regimes	4	≈ 0.29 pp.
4. Derivation of EL2N-SL Score	4	≈ 0.52 pp.
5. Role of teacher strength in SL+KD satura-	4	≈ 0.32 pp.
6. Additional analysis for small-scale methods	5	≈ 0.68 pp.
7. Additional results on DCS	5	≈ 2.43 pp.
8. Additional results on the proposed method	8	≈ 2.03 pp.
9. Image Visualization	10	≈ 5.72 pp.
References	16	≈ 1.95 pp.

• Table 1 · p. 3
• Table 2 · p. 3
• Figure 1 · p. 5
• Table 3 · p. 5
• Table 4 · p. 5
• Figure 2 · p. 6
• Figure 3 · p. 6
• Figure 4 · p. 7
• Table 5 · p. 7
• Table 6 · p. 8
• Table 7 · p. 8
• Table 8 · p. 9
• Figure 5 · p. 9
• Table 9 · p. 9
• Table 10 · p. 10
• Table 11 · p. 10
• Figure 6 · p. 11
• Figure 7 · p. 12
• Figure 8 · p. 13
• Figure 9 · p. 14
• Figure 10 · p. 15

Argument and organization

1. Problem and contribution

   Begins from disagreement between distilled-data results and random/coreset baselines, then isolates the role of teacher-provided labels.

2. Argument sequence

   Defines hard, fixed-soft and online soft-plus-KD regimes before proposing a diagnostic correlation score and a new method.

3. Evidence and verification

   Table 1 holds architecture fixed and repeats IPC columns for three label regimes, with a same-compute full-dataset reference.

Problem, claim and contribution

Begins from disagreement between distilled-data results and random/coreset baselines, then isolates the role of teacher-provided labels.

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

Argument and section order

Defines hard, fixed-soft and online soft-plus-KD regimes before proposing a diagnostic correlation score and a new method.

Sections 2–5, pp. 3–8 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 separates compute scaling, score distribution and hardness response into coordinated panels rather than hiding them in one aggregate curve.

Figure 2, p. 5 · Main paper · Additional source 2

Tables, comparisons and evidence

Table 1 holds architecture fixed and repeats IPC columns for three label regimes, with a same-compute full-dataset reference.

Table 1, p. 4 · Main paper · Additional source 2

Appendix and reproducibility

Teacher-strength, cross-architecture and correlation robustness studies accompany exact synthesis/training settings and example distilled images.

Supplement Sections 2–9, pp. 3–15 · Main paper · Additional source 2

Clarity and writing conventions

Footnotes explicitly qualify the apparent irrelevance of data quality: real-image distribution and a minimum subset size remain assumptions.

Section 3, pp. 4–5 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For empirical re-evaluation, expose the hidden resource or supervision contract before interpreting the leaderboard.

Boundary: Conclusions about weak data-quality effects are regime-specific; they do not mean arbitrary noise works or that hard-label dataset design is unnecessary.

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 1–7, PDF pp. 1–8; references pp. 9–10: abstract/framing, section sequence, figure/table captions, evaluation organization and conclusions; not an exhaustive technical verification.

Appendix / supplement

Writing-oriented supplement scan: Supplement Sections 1–9, pp. 1–17: objectives, hyperparameters, scaling fits, score derivation, teacher strength, transfer, DCS robustness and image examples. Video/code assets were not executed or assessed.

Visually inspected PDF pages

5, 4

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 · 10 pages
• Official CVF supplementary PDF · 17 pages