Guide
Rejected from CVPR: nobody will ask what you changed
None of CVPR, ICCV or ECCV asks for your previous reviews or a change log, so rejected papers go back nearly unchanged. The rules that do bind a resubmission sit elsewhere, and one turns the obvious fix into a violation.
Published 14 August 2026, dates rechecked 10 September 2026
CVPR said no, and nobody downstream will ask what changed. Which is exactly why the next committee finds the same weaknesses unless you find them first. The open windows in vision are on the deadline board
Read this before you touch the paper
- No published guideline asks what changed. That reads as a clean slate and is not one.
- At ECCV, new experiments in the appendix are a violation, not a fix. It counts self-plagiarism as plagiarism and bans supplementary material holding results obtained with an improved version of the method.
- You joined the reviewer pool when you submitted. At CVPR and ECCV, missing that deadline desk-rejects every paper you author.
- The calendar has moved, and this time it has a date on it. ECCV closed on 5 March, thirteen days after CVPR decided. The next room is CVPR 2027 in Seattle: registration 10 November 2026, paper 16 November, and its call says that day will not be changed.
The calendar
Thirteen days last time, and 16 November this time
Compare the three on what a rejected paper has to become, and on when it had to be ready. The room that is open now is CVPR 2027, which closes on 16 November 2026.
| CVPR 2026 | ICCV 2025 | ECCV 2026 | |
|---|---|---|---|
| Main paper | Eight pages including figures and tables, plus reference pages. | Eight pages excluding references. | 14 pages in the LNCS style, plus reference pages. |
| Appendices in the main file | None permitted. | ||
| Supplementary | Moved onto the paper deadline in 2025, 7 March. | One week after the paper, 12 March 2026. No results from an improved version of the method. | |
| Reviews out, rebuttal shut | 22 to 29 January 2026, seven days. One page, anonymous, no external links. | One page. Longer responses are not reviewed. | 2 to 11 May 2026, nine days, then reviewer discussion to 20 May. |
| Decisions | 20 February 2026 | 25 June 2025 | 17 June 2026 |
| The next edition | CVPR 2027 in Seattle, and it is open: registration 10 November 2026, paper 16 November, decisions 25 February 2027. | ICCV 2027, 2 to 8 October, Hong Kong. Still no call: its 2027 page 404s. | Closed 5 March 2026. No later edition announced as of 10 September 2026. |
An empty cell is a question we did not check there. Read 14 August 2026: CVPR dates and author guidelines, ICCV author guidelines and changes, ECCV dates and policies, ICCV 2027 dates from the CVF. Read 10 September 2026: the CVPR 2027 call for papers, with still no ICCV 2027 page and nothing after ECCV 2026. CVPR prints its June dates two ways, 20 to 24 June on the CVF against 20 to 25 June on its own site, so neither is above.
The revision had to exist before you knew you needed it
CVPR decided on 20 February 2026 and ECCV closed on 5 March, registration six days earlier. Thirteen days, our arithmetic on their dates, to turn eight pages into fourteen in a template you have not used, with no appendices in the main file to take the overflow. Not a fortnight's work, so it had to start in the rebuttal week.
What binds you
The rules that decide it are not about resubmission
Nothing in these guidelines governs sending a rejected paper back. Plenty of them govern what you were about to do to it.
| Rule | Where it is published | What it stops you doing |
|---|---|---|
| Prior reviews and a change log | Nowhere. None of the three published author guidelines asks for either. | Nothing, which is how the paper goes back nearly unchanged. |
| Self-plagiarism | ECCV 2026: plagiarism is appropriating words or results without credit, "both verbatim or otherwise. To clarify, this includes self-plagiarism". | Reusing the rejected text uncredited is on the same list as reusing somebody else's. |
| Improved results in the supplement | ECCV 2026: supplementary material "may not include results obtained with an improved version of the method". | The standard post-rejection move. At ECCV the fix is the violation. |
| Two venues at once | ICCV 2025 counts 20 percent or more overlap as the same publication. ECCV 2026's window runs from 13 March 2026 to the decision. | 20 percent is the only published threshold of the three, and it is a definition, not advice. |
| Talking about it in public | ECCV 2026 embargoes the work from 29 January 2026 until final decisions, and allows arXiv posting but not labelling it an ECCV submission. ICCV 2025 does not treat a technical report as prior art. | Post the rejected paper if you like. Do not call it a submission, and say nothing while ECCV holds it. |
The first row is a checked absence, not a ruling: none of the three published author guidelines asks for either, read 14 August 2026 on CVPR 2026, ICCV 2025 and ECCV 2026, which also carry the quotes. Each rule binds the edition beside it.
Nobody asks what changed, so nothing makes you change it
A venue demanding the previous reviews would force you to argue that the paper moved. None of these three does, and the silence reads as permission. The paper goes back to a committee from the same community, some of whom have read it.
The appendix is where an ECCV resubmission goes wrong
The instinct is to run the missing experiment and put it in the supplement, where the page limit does not bite. ECCV forbids exactly that, so the improved method has to become the method of the paper: a rewrite, on the schedule above.
The trap nobody lists
Your reviewing can kill the resubmission
You joined the reviewer pool when you submitted. The penalty each of them publishes for failing at it does not stop at the paper assigned to you.
| CVPR 2026 | ICCV 2025 | ECCV 2026 | |
|---|---|---|---|
| How you were enrolled | "Authors will be automatically added to the reviewer pool unless they are exempted." | "Every author is required to serve as a reviewer (unless they hold another role in ICCV 2025)." | Automatically added unless exempted. |
| Who is let off | Exemptions exist. The terms are not published. | Another role at the conference. | Another ECCV role, those new to computer vision, undergraduate and Master's students, or exceptional circumstances via the Program Chairs. The only published list of the three. |
| Missing the review deadline | "subject to desk rejection of all papers on which they are an author". Reviews were due 12 January 2026. | Due 23 April 2025. | "will face a desk rejection of all papers on which they are an author at the discretion of the PCs". |
| A review judged irresponsible | "Any reviewer whose review is deemed to be 'highly irresponsible' will face a desk rejection of all papers on which they are an author". | Same penalty. The examples: "one-sentence reviews, reviews generated by Large Language Models (LLMs), reviews that are irrelevant to the paper, or reviews that overlook substantial portions of the paper". |
An empty cell is a question we did not check there. Read 14 August 2026: the CVPR 2026 reviewer and author guidelines, the ICCV 2025 reviewer guidelines and the ECCV 2026 reviewer guide. The same sanction elsewhere: desk rejection explained.
The failure mode that has nothing to do with the paper
Everything above is about the manuscript. This penalty ignores it: reviews filed late or filed thin, and the published consequence is desk rejection of every paper you author, this one included. An afternoon in January protects what you sent in November.
No published scale
Aim at the definition of a poster, not at a number
None of the three publishes its rating or confidence scale, so any table of numbers you find is a reconstruction. What is published is what the area chair is told a decision means.
| What | The published wording | What it means for the revision |
|---|---|---|
| Award | "major advances that will heavily impact the field; will be used by many people, create new capabilities, etc." | Not the argument you are having. |
| Oral | "potential to be very significant; worthwhile for the whole community to hear about." | The bar rejected papers get measured against, usually by their own authors. |
| Poster | "incremental steps that expand the sum of the community's knowledge or add bricks to the cathedral of knowledge; papers introducing useful tools; papers of interest to a subcommunity." | The target, and the most useful sentence any of the three publishes. Incremental is in the definition: become a solid brick, not a breakthrough. |
| Reject | "unlikely to be significant or already known. Provide clear evidence." | Two failure modes, and the chair must evidence them. Read your reviews for which one was asserted. |
| The rating and confidence scales | Not published by any of the three. The OpenReview form holding the ICCV version returns 403 to anyone off the committee. | Anyone quoting you a number-to-word mapping got it somewhere they cannot show you. |
| How the chair reads them | ECCV 2026, to its area chairs: "Don't focus too much on the scores. Instead, look carefully at the comments." | A revision aimed at raising a number aims at the wrong thing. |
| Not beating the state of the art | CVPR 2026: "the fact that a proposed method does not exceed the state-of-the-art accuracy on an existing benchmark dataset is not grounds for rejection by itself". | A review rejecting you on this alone was outside its own instructions. |
Categories quoted verbatim from the CVPR 2026 area chair guides, instructions from the CVPR 2026 and ICCV 2025 reviewer guidelines and the ECCV 2026 area chair guidelines, read 14 August 2026. We looked for a rating scale on nine pages across the three venues and on the review form itself, found none, and so print none.
A paper is rejected for being unlikely to matter or for being already known, and the two take opposite answers: one from positioning, one from the related work. To have them stated in your draft before a committee states them, a paid pre-submission review does that part.
No change log is required. The objections travel anyway.
CVPR 2027 closes on 16 November 2026, registration six days earlier, and its call says that day will not be changed. Specialists read the revised draft against the guidelines of the venue you are aiming at now and hand back the objections, with locations, while the file is still open.
Why the rebuttal here is a different instrument: CVPR, ICCV and ECCV compared. Outside vision: after a NeurIPS rejection.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.