Submission guides

CVPR 2027 submission checklist

20 things to check before you submit to CVPR 2027. scime checks all of them on your LaTeX project — page count on the compiled PDF, every reference against arXiv, DBLP, Crossref and OpenAlex — in about a minute.

CVPR 2027 registration deadline: (AoE)7 days left

Check your paper now, no sign-up

Drop the ZIP of your LaTeX project (in Overleaf: Menu → Download → Source). The quick check reads the source for CVPR 2027 rules, anonymity leaks, missing figures and labels, cite keys missing from your .bib, and typos. Checked in memory and not stored; the page count and reference lookups run in the full check after sign-up.

Drop your .zip here

Up to 100 MB · Overleaf: Menu → Download → Source

No ZIP at hand? Try it on our example paper, which has deliberate mistakes.

Key rules at a glance

Main text, submission
8 pages (references and appendix not counted)
Main text, camera-ready
8 pages
Review
Double-blind: no author names, self-identifying links or acknowledgments
Official rules
cvpr.thecvf.com/Conferences/2027/AuthorGuidelines

Rules change between years. The official call for papers always wins; tell us if something here is out of date (contact@latenticlabs.com).

CVPR 2027 rules

  • Main text fits in 8 pages for review

    Measured on the compiled PDF up to the references. Over-length submissions can be desk-rejected.

  • The official CVPR 2027 style file is used, in submission (review) mode

    Submission mode keeps the author names hidden; the camera-ready option shows them.

New for CVPR 2027

  • An “LLM Usage Statement” section after the main text, before the references

    Required; papers without it are desk-rejected. Name the LLM or generative AI tools, what they were used for and how much, or write “No LLMs or generative AI tools were used in this work.” It does not count toward the 8 pages.

  • The CVPR 2027 submission template

    Papers that do not use it are rejected without review; older kits (for example CVPR 2025) still print a paper ID.

  • No tracking, video or project-page links

    Links that expand the paper, track readers or reveal the authors are desk-rejected; put videos and code in the supplementary material.

  • No hidden text aimed at LLM reviewers

    Hidden or visible instructions meant to manipulate reviewers or automated tools (prompt injection) are an ethics violation.

  • No datasets withdrawn by their creators

    For example DukeMTMC-ReID or MS-Celeb-1M; if you must use one, justify it in detail.

  • An IRB and consent statement for user studies

    Say which review board approved or oversaw the study and that consent was obtained, or explain why not.

Anonymity (double-blind review)

  • No first-person references to your own prior work

    Write “Smith et al. (2024) showed…”, not “in our previous work”.

  • Code and model links are anonymized

    A github.com or huggingface.co link to your account reveals who you are; use an anonymous mirror.

  • No acknowledgments or \thanks footnotes

    Funding and affiliation notes identify the authors; add them back for the camera-ready version.

  • The PDF has no author metadata or email addresses

    hyperref’s pdfauthor and stray emails survive in the compiled file.

References

  • Every \cite key has a bibliography entry

    Missing keys print as [?] in the PDF.

  • Each arXiv ID and DOI points to the paper you meant

    Copied BibTeX often carries the ID of a different paper.

  • Every cited paper exists, with the right title and authors

    Each entry is looked up on arXiv, DBLP, Crossref and OpenAlex; AI-suggested references are the usual culprit.

Check your .bib file now, free and without signing up →

Missing items and LaTeX slips

  • Every \ref points to an existing \label

    Broken cross-references print as ?? in the PDF.

  • No leftover TODO, FIXME or highlighted text

    Reviewers notice; so do area chairs.

  • Every figure and table has a caption, a label and a mention in the text

    Unreferenced floats read as unfinished work.

  • No unescaped % after a number

    In LaTeX, 41% starts a comment and silently deletes the rest of the line.

  • No repeated words, a/an slips or mixed spellings

    Cheap to fix, and they shape a reviewer’s first impression.

Run the whole checklist on your paper

Upload your Overleaf ZIP or start from the CVPR 2027 template. The full report is free: every issue with its line, plus AI proofreading. Not ready to upload? Try it on our example paper, which has deliberate mistakes.