Optical Flow
Methods that estimate dense 2D motion between consecutive frames and remain accurate under realistic visual corruptions.
Call for Papers
Robust Correspondence Under Realistic Visual Corruptions
We invite concise workshop papers on robust Optical Flow, Stereo Matching, Scene Flow, and rigorous exploration of robustness in dense correspondence.
Four Tracks
What We Invite
Stereo Matching
Methods that estimate dense disparity from rectified stereo pairs robustly across clean and corrupted imagery.
Scene Flow
Methods that estimate dense 3D motion from stereo video while preserving geometry and temporal accuracy under corruptions.
Exploration Track
Focused, reproducible studies of failure modes, evaluation, uncertainty, adaptation, temporal consistency, or robustness trade-offs.
Submission Requirements
Format and Review
Every quantitative or Exploration Track entry must be accompanied by a workshop paper.
Challenge Teams: Identify Your Entry
Include your exact registered team number and team name in the paper abstract—for example, RoCo-XX · Example Team.
-
Format: Use the official
NeurIPS 2026 paper template.
Select
\usepackage[sglblindworkshop]{neurips_2026}and set\workshoptitle{RoCo-Spring: The Robust Correspondence Challenge}. - Length: The main paper must be 4–6 pages. References and an optional appendix may follow and do not count toward the 4–6-page main-paper limit.
- Review: Submissions are single-blind; author identities remain visible to reviewers.
- Content: Quantitative papers should report the method, data, compute, results, ablations, and limitations. Exploration papers should present a focused question, code, evidence, and reproduction instructions.
Acceptance is based on technical soundness, clarity, reproducibility, and relevance to robust dense correspondence—not leaderboard rank alone.
Camera-Ready
Reproducibility Package
With the camera-ready paper, submit code and concise reproduction instructions that include:
- A declaration listing every training, pretraining, fine-tuning, and validation dataset.
- Runnable training and evaluation scripts.
- GPU type and count, peak memory, wall-clock time, and other compute requirements.
- A README with a pinned environment via
pyproject.tomlor explicit conda packages and versions.
For the camera-ready version, add the template's final option. Award eligibility
requires an accepted paper and a complete reproducibility package.
Key Dates
Paper Timeline
| Date | Milestone |
|---|---|
| September 25, 2026 NEW | Workshop paper submission deadline |
| September 30, 2026 | Final quantitative benchmark submission deadline |
| October 06, 2026 | Author notification |
| October 15, 2026 | Camera-ready paper, code, and reproducibility package deadline |
Dates are tentative and subject to change.