Guide
CVPR vs ICCV vs ECCV: you are choosing between two
Only two of these three exist in any given year. CVPR runs annually, ICCV in odd years and ECCV in even ones, so the three-way decision everyone writes about is not one you ever actually make.
Published 11 August 2026
Already have a CVPR decision? Then the next room is the question, and the calendar decides which of these is open. After a CVPR rejection
The short version
- You are choosing between two, never three. CVPR runs every year, ICCV in odd years and ECCV in even ones, so the three-way comparison everyone writes about is not a choice you ever make.
- In 2026 that means CVPR and ECCV; in 2027, CVPR and ICCV. ICCV ran in 2025 and returns in 2027, with no 2026 edition at all.
- CVPR is the default rather than the best. It is the fixed point you can plan a paper toward, and the one whose rejection costs a full year if you have nothing else lined up.
- The CVPR rebuttal window is seven days. 22 to 29 January 2026, on 16,092 submissions of which about one in four was accepted.
The thing nobody puts in the table
Two of them are never in the same year
Every comparison of these three lines them up side by side as though you could pick any one. Put the actual calendar next to it and the choice collapses.
| 2026 | 2027 | |
|---|---|---|
| CVPR | 5 to 7 June, Denver | 20 to 24 June, Seattle |
| ICCV | Does not run | 2 to 8 October, Hong Kong |
| ECCV | 8 to 12 September, Malmö | Does not run |
Dates from CVPR, ECCV, which states it is biennial on even years, and the Computer Vision Foundation, read August 2026. ICCV ran in 2025 and runs again in 2027, with no 2026 edition, which is what the odd-year pattern looks like from the outside.
CVPR is the fixed point and the other two take turns
That makes CVPR the default rather than the best. It comes round every year with a deadline in November, so it is the one you can plan a paper toward, and the one whose rejection costs you a full year rather than a season if you have nothing else lined up.
The real sequence is CVPR first, then whichever is next
For a paper being written now the order is set by the calendar rather than by ambition: CVPR in November, and then ECCV or ICCV depending on the year you land in. Deciding the second venue before you submit to the first is the part people skip, and it is the part that stops a rejection becoming six idle months.
Geography is a real input here, not a tiebreaker
These three rotate continents deliberately, and unlike the machine learning conferences the location genuinely differs year to year. If travel money, a visa or a teaching term decides whether you can attend, that is a legitimate reason to prefer one edition, and it is checkable a year ahead.
The part that will catch you out
The CVPR rebuttal is a different instrument
If your habits come from NeurIPS or ICLR, this is where they break. The window is a week and you cannot answer an objection by running the experiment.
Seven days, start to finish
CVPR 2026 released reviews on 22 January and closed the rebuttal on 29 January. That is the whole thing. There is no multi-week discussion phase to spread the work over, so a rebuttal has to be planned before the reviews arrive rather than after, and the week has to be clear in your calendar in advance.
No substantial new experiments, by rule
The guidelines let you include figures and comparison tables drawn from results already in your submission, your supplemental material or other published papers. They do not let you answer a missing-experiment objection with a new experiment. This is the single biggest difference from the venues most rebuttal advice is written about, including ours.
So the supplemental material is your rebuttal budget
Anything you might need to point at later has to be in the submission or the supplement on the day you submit. Results you leave out because they seemed peripheral cannot be brought in during the response, and that is the decision people regret in January rather than in November.
The general rebuttal guide tells you to run the missing experiment if the window allows. At CVPR it does not, so read that page with this rule in front of it.
Scale
What CVPR publishes, and what the other two do not
CVPR states its numbers plainly. For 2026 it received 16,092 submissions, a 24% rise on the year before, and accepted 4,089 of them, about one in four.
Neither ICCV nor ECCV publishes an acceptance rate on its own site, and we are not going to supply one. Figures for both circulate widely and the pages carrying them give no source, which is how a number becomes true by repetition. If you need one, it belongs to the programme chairs of the specific edition, and their announcement is the only version worth quoting.
What the CVPR figure is good for is calibration rather than comparison. Three papers in four are turned down at the venue that publishes its numbers, and there is no reason to think the other two are gentler. A rejection here is the ordinary outcome for competent work, which is the same thing the machine learning conferences show. What that means for your own paper is a separate question, and it has an answer.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
Decide the second venue before you need it
The matcher ranks the venues that publish your subject against your abstract, which is the cheapest way to have the fallback picked while you can still think clearly about it. Free, no account, and nothing you paste is kept.
Before the November deadline, the list of things that end a submission without a review is the pre-submission checklist.
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.