Iccv workflow

ICCV Workflow

An ICCV campaign runs on a spring-deadline, autumn-conference rhythm that only exists in odd years. The 2025 cycle — the most recent completed one, verified 2026-07-08 — ran: paper registration March 3, full submission and supplementary March 7 (11:59pm HST), reviews to authors May 9, rebuttal May 10–16, decisions late June, conference October 19–23 in Honolulu (workshops the 19th–20th, main program the 21st–23rd). ICCV 2027 is announced for Hong Kong in early October 2027; its deadline chain was unpublished at check time, so every date below is a 2025-cycle anchor for shape, not a promise.

The autumn fork (the decision nobody schedules)

The ICCV cycle actually begins the previous autumn, when the team must choose between the imminent CVPR deadline (November) and holding four more months for ICCV (March). Make the fork an explicit meeting with a written outcome:

• Hold for ICCV when the missing evidence is real (a benchmark, a user study, a second domain) and four months genuinely closes it.
• Take CVPR when the paper is submittable now — holding a finished paper four months in a field with weekly arXiv turnover is negative-expected-value.
• Never split the difference by submitting to CVPR "as a free review round"; overlapping CVF reviewer pools remember, and dual-submission windows can trap a paper that gets held over.

Cycle plan with owners (2025-shaped)

Window	Milestone	Failure this window causes
Sep–Oct (year before)	Autumn fork decided; claims frozen in one paragraph	Zombie project drifting between deadlines
Nov–Dec	Experiment matrix runs; draft skeleton with figure slots	March becomes an experiments deadline, not a writing one
Jan	Full internal draft; outside reader pass	No time to act on feedback
Feb	Baseline-freshness audit; anonymity sweep; supplement built in parallel	Same-day supplement rule catches the team
~Mar 3	Registration: title, abstract, full author list, complete OpenReview profiles	Registration lapse = no submission
~Mar 7	Paper + supplement, one upload event (2025: 11:59pm HST)	Anything unfinished is unfinished forever
Mar–May	Coauthors serve reviewer assignments — every author reviews at ICCV	Late reviews can desk-reject your paper
~May 9–16	Reviews land; seven-day rebuttal	Week lost to shock instead of triage
Late Jun	Decision	No plan B drafted
Summer	Camera-ready (exact 2025 deadline 待核实); artifact release runway	Promises to reviewers quietly expire
Oct	Conference week abroad	Visa/travel booked too late

Note the timezone trap: 2025's deadline was Hawaii Standard Time — generous for Europe and Asia, but teams that reflexively assume AoE have missed CVF deadlines in both directions. Read the clock on the current Dates page, not from habit.

Two ICCV-specific project risks

The unified deadline. Since 2025, ICCV takes paper and supplementary material on the same day (an announced process change — it also shortened the decision timeline). Plan the supplement as a parallel workstream from February, with its own owner. Teams migrating from CVPR habits still budget a "supplement week" that no longer exists.

Reviewer duty with teeth. Every author is expected to review unless exempt (those new to or outside the vision community were exempted in 2025), each author may register at most 25 submissions, and the 2025 chairs enforced quality: 25 reviewers were judged highly irresponsible and 29 of their authored papers were desk-rejected, 12 of which would otherwise have been accepted. Treat coauthor review assignments as deliverables on the project plan with a senior reviewer proofreading first-timers' drafts.

International logistics are part of the plan

ICCV rotates internationally by design — Honolulu in 2025, Hong Kong announced for 2027. For many team members that means visas, and visa appointment queues in some countries exceed the June-decision-to-October-conference gap. Do not wait for the decision:

travel-readiness ledger (start at submission, not at acceptance)
  per author:  passport expiry ≥ 6 months past conference?  visa type needed?
               appointment backlog in their country (weeks)?
  per paper:   who presents if the first author cannot travel?
               funding source confirmed for at least one presenter?
  deadline:    visa paperwork drafted by decision day, filed within a week of it

The fallback ladder (the biennial part)

Missing ICCV costs two years of ICCV, but not two years of publishing. Draft the ladder before decisions arrive:

1. Reviews were fixable → ECCV (the following even-year spring) or the next CVPR (November): repair every factual objection first; pools overlap.
2. Reviews said "sound but not exciting" → in 2025, the new Findings workshop accepted technically sound rejected/withdrawn ICCV papers (a post-decision outlet — check whether it recurs); or WACV for application-leaning work.
3. Reviews revealed a missing foundation → journal track (TPAMI/IJCV) after real revision, or back to the bench.
4. Idea overtaken during review → salvage the reusable component; do not resubmit a superseded comparison table anywhere.

Multi-paper labs, odd-year edition

In odd years a vision lab may push papers at CVPR (Nov), ICCV (Mar), and ICCV workshops (summer CFPs) with the same people. Lab-level coordination: one shared duty roster (a single delinquent reviewer-coauthor endangers every ICCV paper the lab submitted), one shared compute reservation for rebuttal week in May, and one person owning OpenReview profile hygiene across all submissions — the 2025 cycle required complete profiles at registration.

Reverify before trusting any date here

• The entire 2027 chain: registration/paper deadlines, timezone, rebuttal window, decision date, camera-ready — all 待核实 until iccv.thecvf.com posts them.
• Whether the single paper+supplement deadline persists.
• Whether Findings runs again in 2027.
• Registration and presentation obligations for accepted papers (not verified even for 2025 — 待核实).

Output format

[Cycle position] autumn-fork / building / deadline-month / in-review / rebuttal / decided / pre-conference
[Next 3 milestones] <date — item — owner>
[Duty roster] coauthor reviewer assignments tracked: yes/no
[Supplement workstream] parallel owner assigned: yes/no
[Travel readiness] visas/funding per presenter: green/amber/red
[Fallback ladder] <ranked next venues with their real deadlines>