brycewang-stanford ECCV Author Response
Use this when ECCV reviews land. In the 2026 cycle reviews were released May 2, the rebuttal was due May 11, and reviewers received rebuttals on May 12 — nine calendar days to answer everything. The rebuttal was optional, a single-page PDF including references, set in a two-column rebuttal template even though the paper itself is single-column LNCS.
What the one page is for
The rebuttal's real reader is the area chair running the post-rebuttal discussion, not the reviewer who wrote the complaint. Write each block so an AC can lift it directly into a meta-review: objection, answer, pointer into the submitted PDF.
- Answer misreadings first — they are the cheapest score to recover.
- Answer requests whose evidence already exists in the paper or supplement second; cite figure/table/line numbers, never "see above".
- Concede real limitations in one sentence each; a clean concession reads better in discussion than a strained defense.
- Skip philosophical disagreements about importance; one page cannot win them.
Budgeting a page against three reviews
| Block | Share of the page | Content rule |
|---|---|---|
| Header + shared clarification | ~15% | One paragraph fixing the misunderstanding all reviewers share, if any |
| Reviewer 1 (most negative, fixable) | ~35% | Two or three numbered points, each ending in a PDF pointer |
| Reviewer 2 | ~25% | Same pattern, shorter |
| Reviewer 3 (already positive) | ~10% | Thank, confirm, correct one detail |
| References | ~15% | Only citations actually used in the rebuttal — they count against the page |
References counting inside the single page is the ECCV-specific squeeze: every citation you add deletes a sentence of argument. Cite by number into the paper's own bibliography where possible instead of restating entries.
New-experiment judgment
ECCV rebuttal norms on new results vary by cycle and are enforced by ACs, so check the current author FAQ before pasting fresh numbers (待核实 for any cycle you are in). The safe pattern regardless of cycle:
- If a number already exists in the supplement, quote it — that is retrieval, not new work.
- If a small experiment is genuinely decisive and the FAQ permits it, report it in one row with its protocol named, no new figure.
- Never promise a camera-ready experiment you have not already run; ECCV ACs read unbacked promises as concessions.
Nine-day schedule
Day 0-1 (May 2-3): read all reviews twice; classify every point:
misread / evidence-exists / real-gap / opinion
Day 2-3: draft answers for misread + evidence-exists points
Day 4-5: run at most ONE decisive check if the cycle allows it
Day 6: compress to one page in the official rebuttal template
Day 7: co-author pass — every claim gets a PDF pointer or dies
Day 8 (May 10): freeze; upload a day before the May 11 cutoff
Tone calibration for a European generalist panel
ECCV panels mix subfields more than a specialist workshop would; the reviewer who scored you lowest may be adjacent to, not inside, your niche. Spell out what a neighboring-subfield reader needs (what the benchmark measures, why the baseline is the right one) instead of assuming shared folklore. Avoid sarcasm and priority fights — the AC discussion happens in your absence, and your prose is the only advocate present.
Output format
[Rebuttal verdict] worth submitting / skip (already sunk or already safe)
[Point map] <review point -> misread / evidence-exists / real-gap / opinion>
[Page budget] <block -> lines allocated>
[PDF pointers] <claim -> figure/table/section in the submission>
[Risks] <new-result rule check, reference overflow, tone flags>