Writing study: Differentiable Vector Quantization for Rate-Distortion Optimization of Generative Image Compression

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

SHIYIN JIANG, Wei Long, Minghao Han, Zhenghao Chen, Ce Zhu, Shuhang Gu · 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

≈ 8.1 page-equivalents

Main figures / tables

10 / 1

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 73.64%References / intervening material: ≈ 26.36%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	≈ 1.36 pp.	12.36%
2. Related work	2	≈ 0.99 pp.	9%
3. Methodology	3	≈ 2.19 pp.	19.91%
4. Experiments	5	≈ 2.94 pp.	26.73%
5. Conclusion	8	≈ 0.27 pp.	2.45%
References	9	≈ 2.9 pp.	26.36%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 4
• Figure 4 · p. 5
• Figure 5 · p. 6
• Figure 6 · p. 6
• Figure 7 · p. 7
• Figure 8 · p. 7
• Table 1 · p. 7
• Figure 10 · p. 7
• Figure 9 · 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 12 physical pages; 14 detected figures and 1 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 11.93 page-equivalents, 14 figure labels and 1 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.02 pp.
A. Architecture Details	1	≈ 1.03 pp.
B. Training Details and Hyperparameters	2	≈ 1.08 pp.
C. Experiments	3	≈ 9.8 pp.

• Figure S1 · p. 3
• Figure S2 · p. 3
• Table S1 · p. 3
• Figure S3 · p. 4
• Figure S4 · p. 4
• Figure S5 · p. 4
• Figure S6 · p. 5
• Figure S7 · p. 5
• Figure S8 · p. 6
• Figure S9 · p. 7
• Figure S10 · p. 8
• Figure S11 · p. 9
• Figure S12 · p. 10
• Figure S13 · p. 11
• Figure S14 · p. 12

Argument and organization

1. Diagnose the missing gradient

   Contrast scalar and vector quantization before the rate–distortion formulation.

2. Define implementation and evidence

   Explain soft rate relaxation and causal entropy prediction, then report curves, visual crops and ablations.

Problem, claim and contribution

Figure 1 locates the claimed bottleneck at the stopped rate-loss gradient rather than merely saying existing codecs are inefficient.

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

Argument and section order

The method begins with the conventional rate–distortion contract, then explains precisely which branch becomes soft while reconstruction and coding remain hard.

Sections 3.1–3.3, PDF pp. 3–4 · Main paper · Additional source 2

Figures, captions and visual explanation

The overview uses a separate right-hand inset for the VQ operation and dashed colored gradient arrows. This separates forward decoding from the training-only rate pathway without forcing all detail into one box chain.

Figure 2, PDF p. 3; visual inspection · Main paper · Additional source 2

Tables, comparisons and evidence

The page-6 crop grid prints bpp and DISTS beneath every reconstruction, and the adjacent efficiency plot encodes parameters, quality and latency. Quality examples therefore retain their rate/resource context.

Figures 5–6, PDF p. 6; visual inspection · Main paper · Additional source 2

Appendix and reproducibility

Architecture, staged optimization and third-party model evaluation precede the long high-resolution gallery, keeping reproduction settings distinct from visual demonstrations.

Supplement Sections A–C, PDF pp. 1–4; Figures S6–S14, pp. 5–12 · Main paper · Additional source 2

Clarity and writing conventions

The text explicitly says FID is omitted on Kodak because of its small sample count, an economical example of explaining a missing result instead of silently dropping a metric.

Evaluation Metrics, PDF p. 6 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: When optimizing a trade-off, show both axes beside every visual comparison, and draw the gradient path if differentiability is the key contribution.

Boundary: A perceptual-metric win at very low bitrate is not interchangeable with lossless fidelity or superiority at all rates; latency and parameter count also measure different costs.

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 PDF pp. 1–11: framing, argument sequence, section organization, figure/table captions, evaluation interpretation and limitations; reference sections identified. This is a writing-focused scan, not an exhaustive technical proof or reproduction audit. Main PDF pages 3 and 6 were rendered and visually inspected.

Appendix / supplement

Writing-focused scan of the separate supplementary PDF, pp. 1–12, including its organization, captions, protocol/evidence detail and limitation material where present; proofs and code listings were inspected for exposition, not independently verified or executed. Supplement Sections A–C, PDF pp. 1–4; Figures S6–S14, pp. 5–12. Linked video/audio demonstrations and non-PDF archive contents were not assessed.

Visually inspected PDF pages

3, 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 CVPR 2026 CVF open-access accepted manuscript; retrieved 2026-09-26 · 11 pages
• Official CVPR 2026 CVF supplementary PDF; retrieved 2026-09-26 · 12 pages