brycewang-stanford WACV Writing Style
Use this to make a WACV paper read the way its reviewers judge it. The governing fact is the two-round model: a first page that answers the reviewer's obvious question earns Accept; one that leaves it open earns Revise and Resubmit. Facts are the WACV 2026/2027 cycles as read on 2026-07-09.
Write for your track
The two tracks reward different first pages, and writing to the wrong one draws the wrong review:
| If your track is… | Lead the first page with… | The reviewer is asking… |
|---|---|---|
| Applications | The real-world constraint and why existing methods break under it | Does this system actually work under the deployment constraint you name? |
| Algorithms | The method/mechanism and the matched-baseline comparison | Is the algorithmic idea new, and does the gain survive fair comparison? |
An Applications paper that opens with "a novel attention module" and an Algorithms paper that opens with "an important real-world problem" have each buried their contribution under the other track's framing.
The first-page arc
Put the contribution in the first page: problem/constraint → why prior methods are insufficient (each named) → method or system → evidence → why it fits the track. Every major claim on that page should point to a figure, a table, or a section — not to a promise. Because WACV reviewers see three reviews reconciled by an AC, a page that a busy reader can summarize in two sentences travels further than a dense one.
Figures are page budget
The cap is 8 pages including figures and tables. A teaser, two qualitative grids, and three result tables can eat half the paper, so design figures as line items:
- One teaser that states the contribution visually, with a self-contained caption a reviewer can read without the body.
- Qualitative results that a reviewer will judge by eye — check they render at print zoom.
- Tables that report the comparison under the stated condition, with uncertainty.
Caption test (apply to every figure):
Can a reviewer who reads ONLY this caption state (a) what is shown and
(b) what it demonstrates about the claim? If not, rewrite the caption, not the body.
Scope every claim
WACV's applications framing punishes overclaiming quickly, because a reviewer can often name the deployment case you did not test. Scope claims to the domain, constraint, and hardware you actually demonstrated ("row-crop segmentation at 2 W in sub-10-lux light"), and disclaim what you did not show. A scoped claim that holds beats a broad claim a reviewer can break in one sentence — and broken claims are what turn into revision requests.
Compress for eight pages
- Move protocol detail, extra qualitative results, and derivations to the anonymized
supplement (see
wacv-supplementary), keeping the body's argument intact. - Cut roadmap sentences that restate the table of contents; spend the space on the gap paragraph, where reviewers form their opinion.
- Keep notation and baselines consistent between body and supplement so a Round 2 reviewer re-reading a revised paper is never confused by a moved number.
Reverify each cycle
- The 8-page-including-figures rule and the references-only overflow.
- The two-track review criteria and any track-specific formatting.
- The author-kit style version.
Output format
[Track fit] first page leads with the right thing for <Applications|Algorithms>: yes/no
[First-page arc] constraint/method → gap → evidence present: yes/no
[Figure budget] pages spent on figures/tables: <n of 8> · captions self-contained: yes/no
[Overclaim risk] <claim a reviewer could break> → scoped: yes/no
[Compression] what moves to supplement: <list>