CVPR 2026 Paper Formatting Template

Submission guidelines

CVPR is a top computer vision conference. Reviewers care about clear problem statements, strong experimental comparisons, and trustworthy visualizations. CVPR has long used the official CVF template; two-column measure, 8-page main text, and consistent IEEE citations matter. Format issues rarely reject a paper alone but hurt professional perception and distract reviewers during rebuttal.

Vision papers are often rejected for weak baselines, insufficient ablations, or weak generalization arguments. Format risks: figures whose text is unreadable after two-column scaling, spanning figures that break pagination, and mismatched citation numbers vs. the bibliography.

Formatting in detail

• Page and layout: US Letter; two columns; official CVF format; 8 pages maximum for main text; references usually do not count toward main text (per that year’s guide).
• Margins: 1 inch top and bottom; 0.75 inch left and right.
• Fonts: body Times New Roman 10pt; headings 14pt (as in cvpr template).
• Required structure: Abstract, Introduction, References; methods, experiments, and related work follow community practice and template compatibility.
• Authoritative source: CVPR 2026 Author Guidelines.

Abstract guidelines

• Length: typically 200 words maximum.
• Content: task and data, method summary, main quantitative/qualitative results; you may name metrics (e.g., mAP, FID) without number overload.
• Style: concise and self-contained; avoid figure references; expand acronyms on first use.
• Integrity: claim “SOTA” only when rigorously true under the same experimental setting.

Citations and references

• Style: IEEE, numbered citations [1], bibliography in order of first appearance.
• BibTeX: use the official .bst; complete @inproceedings fields; consistent venue abbreviations and page ranges.
• Common mistakes: hand-renumbering citations; mixing \cite styles; duplicate bibliography entries for the same paper.

Frequent formatting mistakes

• Main text over 8 pages: appendix content in the main PDF—use official supplementary material.
• Non-CVF template: old-year class files or modified documentclass options.
• Unreadable figures in two columns: too many subpanels or fonts below readable thresholds.
• Wrong paper or margins: A4 or margins inconsistent with CVF.
• Reference order/number mismatches: insufficient compilation passes or .bib key conflicts.

How to convert to CVPR format with AutoSCI

• Upload your paper — PDF or Word.
• Choose a template — CVPR 2026.
• Export — LaTeX, PDF, or Word.

AutoSCI applies page limits, margins, fonts, and IEEE citation formatting to align with CVPR expectations.

Two-column layout and vision figures

CVPR papers rely heavily on figures. Choosing single-column vs. figure* spanning figures affects page flow and readability: wide floats can create awkward blanks on odd/even pages. Put primary results in a clear spanning figure when needed; use side-by-side single-column layouts for secondary results. All text in figures should be readable at 100% PDF zoom—blurriness after zooming frustrates reviewers.

Color and print

For heatmaps and attention maps that rely on color, ensure grayscale readability—add contours or peak markers. Color-blind-friendly palettes are increasingly expected in the vision community.

LaTeX and the CVF class

Use official cvpr.cls and example .tex files; avoid packages that change the text block. For rebuttal line numbers, use only lineno setups allowed by the venue so they do not clash with two-column mode.

Anonymity and datasets

Under double-blind review, strip author-identifying information; for private datasets, describe acquisition and use compliance. Cite public datasets via their papers, not URLs alone.

Experiments and statistics

Report mean and variance across random seeds; in comparison tables, align training budget and input resolution. If you use significance tests, document assumptions and test types in the appendix.

Pre-deadline checklist

• US Letter, two columns, CVF template.
• Main text ≤8 pages (per that year’s definition).
• Margins 1 in (vertical) / 0.75 in (horizontal).
• Abstract ≤200 words.
• IEEE numbered citations match the bibliography order.
• Figures are sharp in column width; prefer vector graphics.

Tips for clear English and figures

• Writing: unify terminology in methods; make captions self-contained so they do not only say “see Figure X.”
• Deadlines: submit before the official AoE cutoff with margin.
• Licensing: verify licenses for third-party assets and pretrained weights and disclose them in the paper.

Division of labor: main paper vs. supplementary

Keep core proofs and key experiments in the main paper; move extra ablations, visualizations, and implementation details to supplementary material. Number supplementary figures/tables consistently with in-text references—avoid “see supp. Figure 7” without a shared numbering scheme referenced in the main text.

Additional FAQ

A: Use the official two-column CVF template unless the organizers announce an exception.

A: Do not modify the template; cut main-text content or adjust figure layout instead.

A: Prefer published venue, volume, and pages when available; for arXiv-only papers, follow IEEE electronic-entry conventions.

Related work and introduction

Organize related work by task line or method family, not chronology only. End the introduction with explicit contributions—distinct from the abstract, which emphasizes results; the introduction emphasizes motivation and positioning.

Math and pseudocode

Keep pseudocode within single-column width; break long formulas with aligned. Keep notation consistent between methods and figure legends.

Workflow with AutoSCI

Migrating from NeurIPS/ICLR (often APA-like author–year) to CVPR (IEEE) requires replacing the entire citation system and rebuilding bibliography order. Freeze content before template conversion to reduce numbering errors.

Reviews and rebuttal

New figures in rebuttal must still respect page and anonymity rules. When addressing format comments, cite before/after page numbers. Avoid “we fixed it” without locations.

Ethics and data

For faces, biometrics, or misuse-prone techniques, follow CVPR ethics guidance; data collection must comply with regulations and informed consent where applicable.

Video and demo materials (if requested)

If the conference invites video abstracts, respect duration and resolution limits; do not leak identity under anonymous review.

Version notice

The CVPR 2026 Author Guidelines and CVF templates are authoritative; rules may change year to year.

Print preview and font embedding

Before submission, verify embedded fonts in the PDF; avoid orphaned section titles at the end of two-column pages. Print preview with US Letter, actual size.

Team split

One person owns figure export (DPI, line width, font embedding); another owns .bib cleanup to reduce merge conflicts.

Migrating from journal format

From IEEE TPAMI and similar journals, note tighter page limits and different two-column widths—recrop and rearrange subfigures.

Common figure mistakes

• Axes without units or misleading axis ranges.
• Color bars disproportionate to the main image.
• Method names inconsistent between figures and tables.

Final search before submit

Search for: TODO, ??, Figure ??, author emails and affiliations (anonymous phase). Submit only after a clean pass.