展开原文（8437 字）
OVERVIEW
RoCo-Spring: The Robust Correspondence Challenge
A NeurIPS 2026 Challenge on robust dense correspondence under realistic distribution shifts, covering optical flow, stereo matching, and scene flow.
Dense correspondence has made remarkable progress on standard benchmarks. Although modern optical flow, stereo matching, and scene flow methods achieve strong performance under ideal conditions, their accuracy can degrade substantially under real-world visual shifts.
Robustness under realistic corruptions remains a challenge for the deployment of these methods in autonomous driving, robotics, and other safety-critical settings. RoCo-Spring jointly evaluates clean accuracy and corrupted robustness, targeting methods that remain accurate under camera noise, adverse weather conditions, blur, compression, and changes in illumination.
Participate
OpenReview
Submission
Evaluation
Starter kit (codebase)
Tracks
Optical Flow · Stereo · Scene Flow · Exploration
Registration
Register or manage your team
Leaderboards
Coming soon
DATASETS
Spring and RobustSpring
Spring
Spring provides high-resolution stereo video with left and right images at 1920x1080 px, together with dense scene-flow ground truth at 4x super-resolution. It includes bidirectional disparities for stereo and depth, disparity changes for scene flow, and forward/backward optical flow for both left and right views. This makes Spring a unified high-detail benchmark for optical flow, stereo matching, and scene flow.
Paper
Benchmark
Dataset
RobustSpring
RobustSpring extends Spring into a robustness benchmark with 20 realistic image corruptions, including blur, color changes, noise, quality degradations, and weather effects. The corruptions are applied to stereo video data and are integrated consistently over time, across stereo views, and with depth where applicable, enabling controlled robustness evaluation for optical flow, stereo, and scene flow.
Paper
Benchmark
Dataset
TRACKS
Four challenge tracks
Optical Flow
Estimate a dense 2D displacement field between consecutive frames, assigning each visible pixel a horizontal and vertical motion vector.
Stereo Matching
Estimate a dense disparity map from a rectified stereo image pair, assigning each pixel the horizontal offset to its corresponding point.
Scene Flow
Estimate dense 3D motion from stereo image sequences, combining per-pixel geometry and temporal motion into a single correspondence task.
Exploration Track
Submit a rigorous analysis of robustness in dense correspondence, focusing on failure modes, method behavior, metrics, or evaluation design.
TIMELINE
Competition schedule
DATE MILESTONE
July 2, 2026 Website launch
July 13, 2026 Development leaderboard opens for all quantitative tasks
September 15, 2026
NEXT UP
4–6 page workshop paper submission deadline
September 29, 2026 Workshop paper author notification
September 30, 2026 Final quantitative submission deadline
October 7, 2026 Camera-ready paper, code, and reproducibility package deadline
October 15, 2026 Final evaluation, reproducibility checks, and award shortlisting
October 31, 2026 Winners notified and workshop program finalized
December 11 or 12, 2026 In-person NeurIPS Competition Track workshop
All dates are tentative and subject to change.
WORKSHOP
Confirmed keynote speakers
Jia Deng
Princeton University · Princeton Vision & Learning Lab
Jia Deng is Professor of Computer Science at Princeton University and Director of the Princeton Vision & Learning Lab. He brings a uniquely relevant perspective to RoCo-Spring: his work spans benchmark-driven computer vision, large-scale datasets such as ImageNet[1], and influential dense correspondence architectures across multiple tasks, including RAFT for optical flow[2], RAFT-Stereo for stereo matching[3], and SEA-RAFT for efficient and accurate optical flow[4]. His keynote will connect the evolution of correspondence methods and benchmark design with the central question of the challenge: how to turn clean benchmark progress into robust optical flow, stereo, and scene flow methods that remain reliable under realistic visual shifts.
Fatih Porikli
Qualcomm AI Research
Fatih Porikli is an IEEE Fellow and Vice President of Technology at Qualcomm AI Research. His expertise connects robust visual perception, video understanding, low-level vision, and real-world deployment. His prior work includes scene-flow estimation under degraded video conditions, including simultaneous stereo video deblurring and scene flow estimation[5], joint stereo video deblurring, scene flow estimation, and moving-object segmentation[6], and adversarial robustness for vision systems[7]. His talk will provide an industry-facing and deployment-oriented perspective on robust dense correspondence.
REFERENCES
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A Large-Scale Hierarchical Image Database. Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 248–255.
Teed, Z., & Deng, J. (2020). RAFT: Recurrent All-Pairs Field Transforms for Optical Flow. Proc. European Conference on Computer Vision (ECCV), 402–419.
Lipson, L., Teed, Z., & Deng, J. (2021). RAFT-Stereo: Multilevel Recurrent Field Transforms for Stereo Matching. Proc. International Conference on 3D Vision (3DV), 218–227.
Wang, Y., Lipson, L., & Deng, J. (2024). SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow. Proc. European Conference on Computer Vision (ECCV), 36–54.
Pan, L., Dai, Y., Liu, M., & Porikli, F. (2017). Simultaneous Stereo Video Deblurring and Scene Flow Estimation. Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 6987–6996.
Pan, L., Dai, Y., Liu, M., Porikli, F., & Pan, Q. (2020). Joint Stereo Video Deblurring, Scene Flow Estimation and Moving Object Segmentation. IEEE Transactions on Image Processing (TIP), 29, 1748–1761.
Naseer, M., Khan, S., Hayat, M., Khan, F. S., & Porikli, F. (2020). A Self-Supervised Approach for Adversarial Robustness. Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 262–271.
SCHEDULE
Tentative workshop schedule
TIME SESSION FORMAT
08:15–08:25 Opening and challenge overview Organizers
08:25–09:10 Keynote: Jia Deng 35 min talk + 10 min discussion
09:10–09:55 Keynote: Fatih Porikli 35 min talk + 10 min discussion
09:55–10:35 Optical flow track talks 2 talks, 15+5 min each
10:35–11:05 Coffee break 30 min
11:05–11:45 Stereo matching track talks 2 talks, 15+5 min each
11:45–12:25 Scene flow track talks 2 talks, 15+5 min each
12:25–13:25 Lunch 60 min
13:00–15:00 Poster session 2 hours, overlaps with lunch
15:00–15:30 Coffee break 30 min
15:30–16:10 Exploration track talks 2 talks, 15+5 min each
16:10–16:50 Panel and audience forum Organizers, speakers, teams
16:50–17:00 Awards and closing remarks Organizers
Tentative one-day workshop schedule, 8:15 to 17:00.
ORGANIZERS
Organizing team
Shashank Agnihotri
University of Mannheim
Victor Oei
University of Stuttgart
Jenny Schmalfuss
NVIDIA
Katrin Bauer
University of Stuttgart
Henrique Morimitsu
University of Science and Technology Beijing
Andrés Bruhn
University of Stuttgart
Margret Keuper
University of Mannheim and MPI-INF
SUPPORT
Sponsors & acknowledgements
EVENT SUPPORT
Sponsors
This event is supported by the SFB-TRR 161 Quantitative Methods for Visual Computing.
RESEARCH FUNDING
Acknowledgements
Shashank Agnihotri and Margret Keuper acknowledge funding by the DFG Research Unit 5336 – Learning to Sense (L2S). Margret Keuper acknowledges funding by BMFTR project TrackOpt (01IS24074A-D). Andres Bruhn and Victor Oei acknowledge funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 251654672 – TRR 161: Quantitative Methods for Visual Computing (B04, A07). Katrin Bauer and Andres Bruhn acknowledge funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 533085500 – Robust Optical Flow. Katrin Bauer and Victor Oei acknowledge support from the International Max Planck Research School for Intelligent Systems (IMPRS-IS).
German Research Foundation
Learning to Sense
SFB-TRR 161
IMPRS-IS
CONTACT
Get in touch
Challenge email: roco-spring-org@googlegroups.com
Starter kit codebase: hmorimitsu/roco-spring-devkit