Rules & FAQ — RoCo-Spring

Rules

Competition Rules

Each team must register once and use exactly one team ID.

Each submitted archive must follow the task format and include the team ID.

Every challenge-team paper must include the exact registered team name and team number in its abstract.

Test ground truth is private.

Participants must not infer, reconstruct, scrape, or obtain test labels.

All training data, pretraining data, model checkpoints, and external resources must be declared.

External public data is allowed but flagged separately in leaderboard analysis.

Test-time adaptation is allowed only without test labels, manual annotation, or information from future leaderboard evaluations.

Manual per-sample editing of test predictions is prohibited.

Public leaderboard submissions will be rate limited.

Final submissions are frozen after the deadline.

Award candidates must have an accepted workshop paper and submit a complete reproducibility package (training and evaluation scripts, dataset declaration, resource requirements, and environment setup). See the reproducibility requirements for details.

Rule Changes

Rule changes will be versioned, dated, and announced to registered teams.

Workshop

Workshop and Review

Use the official NeurIPS 2026 paper format.
Main papers are 4–6 pages; references and an optional appendix do not count toward that limit.
Reviewing is single-blind, so author identities remain visible to reviewers. Use the template's sglblindworkshop option.
Challenge teams must put their exact registered team name and team number in the abstract.
Acceptance depends on technical soundness, clarity, reproducibility, and relevance to robust dense correspondence, not only leaderboard rank.
Accepted papers may be forwarded for the official PMLR volume of NeurIPS workshop proceedings.

FAQ

Frequently Asked Questions

Can I participate in only one task?

Yes. Single-task participation is allowed. A Cross-Task Leaderboard will rank teams that submit to all three quantitative tasks.

Is external data allowed?

Yes, if it is public and declared. Entries using external data will be flagged separately.

Is test-time adaptation allowed?

Yes, but only without test labels, manual annotation, or future leaderboard information.

Do I need to submit a paper?

Yes. Each quantitative or Exploration Track entry must submit a single-blind workshop paper in the official NeurIPS 2026 format. The main paper is 4–6 pages; references and an optional appendix may follow. Challenge teams must include their exact registered team name and team number in the abstract. See the Call for Papers.

What must the reproducibility package include?

With the camera-ready deadline, each entry must submit training and evaluation scripts, a complete dataset declaration, documented resource requirements, and a README with a pinned environment—either via pyproject.toml or explicit conda instructions with library versions. See the reproducibility requirements for the full checklist.

How are ties broken?

First by the normalized RobustSpring term, then by the normalized Spring term.

Where should questions be asked?

Through GitHub issues and public question threads, or the contact email roco-spring-org@googlegroups.com.

Contact

Get in Touch

Challenge email: roco-spring-org@googlegroups.com

GitHub issues: Report a website or challenge issue