| name | cvpr-paper |
| description | Use this skill to help write, structure, polish, format, or review a paper for CVPR (IEEE/CVF Conference on Computer Vision and Pattern Recognition, any year), or whenever the user mentions 'CVPR', a CVF computer vision submission, the cvpr two-column style, supplementary material for vision papers, the Findings Track, or the CVPR double-blind/rebuttal process. Helps with ideation and narrative, section structure, sentence-level polishing, CVPR LaTeX formatting, figures and tables (including teaser design and leaving sized placeholders for figures), citation verification, rebuttal preparation, and a submission checklist. Double-blind venue. Verify exact dates, page limits, and policies against that year's official CVPR Call for Papers. |
CVPR Paper Writing
Help the user take a computer-vision paper from idea to a polished, correctly formatted,
double-blind CVPR submission, and through the rebuttal phase. You can brainstorm the
narrative, structure sections, polish prose, build LaTeX/figures/tables (or leave sized
placeholders, including the teaser), verify citations, and run a reviewer self-check.
How to use this skill
Load the matching reference file on demand (not all at once):
| User wants | Do this | Reference |
|---|
| Find the story / contribution / outline | nail the one-sentence contribution, then outline | references/ideation-and-structure.md |
| Draft or restructure a section | use the section blueprint | references/ideation-and-structure.md |
| Polish / tighten / "does this flow" | apply clarity principles + claim-evidence map | references/writing-and-polishing.md |
| Figures, teaser, tables, or placeholders | design or reserve space; generate plots | references/figures-and-tables.md, assets/ |
| Generate a flowchart / pipeline / experiment plot | run the helper scripts | references/generating-diagrams.md, assets/make_flowchart.py, assets/make_experiment_figures.py |
| Add citations | verify before citing; never hallucinate | references/citations.md |
| Set up LaTeX / fix formatting / convert venue | template workflow + pitfalls | references/latex-and-submission.md |
| Final pass before submitting | run the checklist below + 5-dimension self-review | this file + references/writing-and-polishing.md |
Be proactive: if the contribution and results are clear, deliver a full draft and flag
open choices, rather than asking permission section by section.
At a glance
- Conference: IEEE/CVF Conference on Computer Vision and Pattern Recognition (annual).
- Submission system: OpenReview / CVPR site (per cycle).
- Review model: double-blind — anonymize the submission, including obvious self-citations.
- Supplementary material is allowed (its own, slightly later deadline) but the main claims
must be understandable without it.
- Recent cycles added a Findings Track for technically sound but more incremental work;
check whether it runs in the target year.
Dates change every year — check the current CFP
Do not rely on memorized dates. For the target cycle, open https://cvpr.thecvf.com and read
off the paper deadline, the separate supplementary deadline (usually about a week later),
the rebuttal window, notification, camera-ready, and conference dates/location. Rough rhythm
(planning only): paper deadline around mid-November of the prior year; conference the following June.
Format (CVPR-specific)
- Template: official CVPR/CVF style (
cvpr.sty), two columns, 10pt, US Letter.
- Length: 8 pages of content excluding references; references are unlimited and do not count.
Confirm the exact rule in the CFP.
Structure notes specific to CVPR
Follow the general blueprint in references/ideation-and-structure.md, with vision emphasis:
- A strong teaser (Figure 1) is critical — design it early or reserve it with a placeholder
(
references/figures-and-tables.md, assets/figure-placeholders.tex).
- Compare against current SOTA on the expected benchmarks for the subfield; missing standard
baselines triggers rejection.
- Ablations must isolate each component's contribution.
- Include clear qualitative result figures; keep claims supported by the main paper, since
reviewers are not obligated to read the supplementary.
Rebuttal phase
- Plan time for the rebuttal; reviewers often request a specific extra experiment or table.
- Answer each point concisely and back it with a number or figure where possible.
Submission checklist (CVPR)
When helping the user
- Ask the subfield (detection, segmentation, generation, 3D, video, multimodal) and target benchmarks.
- Offer to scaffold the CVPR LaTeX skeleton, design the teaser/figure plan (placeholders ok),
build comparison tables, organize main vs supplementary, or draft rebuttal responses.
