Call for Papers: Special Issue on Robust Vision
- Full paper submission deadline: extended to December 16th, 2022
- Review deadline: extended to March 17th, 2023
- Author response deadline: extended to March 24th, 2023
- Final notification: extended to April 14th, 2023
- Final manuscript submission: extended to May 12th, 2023
Oliver Zendel, Austrian Institute of Technology
Adam Kortylewski, Max Planck Institute for Informatics
Torsten Sattler, Czech Technical University
Deep learning has shifted much of the scientific focus in computer vision from algorithmic design towards data design. These changes have led to impressive advances. However, academic advancements are often not translating to equivalent advances for real-world usage. Solutions which perform well in benchmarks and experiments fail to work when applied to the field. This is partially due to the gamification brought in from comparing results on specific datasets. The progress mimics solving of the dataset itself rather than solving the actual underlying task at hand.
In addition to this, the average dataset used during training or evaluation focuses on the average (“easy”) cases to support a strong baseline functionality. Our community has been highly successful in the last years at solving hard computer vision problems for these average cases. Improving the robustness for border cases on the fringe of the system’s requirements is still a mostly unsolved problem. Performance and robustness of systems at rare out-of-distribution (OOD) cases is important for safety and can allow more insights into existing shortcomings. While humans have a broad general basic knowledge to cope with extraordinary situations, most learned solutions will fail to even detect such cases let alone react correctly or safely.
The creation of the “Robust Vision Challenge” workshop and challenge series has been motivated by this gap of usefulness where much perceived progress does not translate into equivalent value for our society.
This “Robust Vision” IJCV special issue shall serve as a catalyst and publication venue for improving our understanding of the underlying problems (dataset bias, generalization, OOD) and also provide a place to discuss approaches to apply and refine academic results to achieve working real-world solutions. Participants of the workshop and scientists only interested in this journal are both invited to submit their papers.
Aims & Scope
Track 1: Robust Vision
- Robustness of Computer Vision Solutions
- General concepts to improve robustness of computer vision solutions
- Real-world applicability vs. academic progress
- Identifying and handling data gaps between training time and application
- Dataset bias
- Impact of dataset bias on training and fair evaluation
- Methods to measure or estimate dataset bias
- Potential bias introduced by validation dataset
- Out-of-Distribution handling
- Detection and management of OOD cases
- Meaningful confidence measures
- Comparison of OOD performance vs. in-distribution cases
- Dataset Design and Combining Datasets
- Design principles for datasets with robustness in mind
- Efficient combination of multiple datasets
- Unification of label policies
- Fair Evaluation Methods
- Approaches to estimate metrics without dataset bias
- Metrics which measure performance at border cases and OOD
- Fair combination and extrapolation of metrics
- Ways to future-proof learned solutions and extend applicability of systems outside their initial specifications
- Solving problems at the end of the long-tail distribution
- Methods to extend usefulness beyond the original scope of a dataset
Track 2: Invite only- Robust Vision Challenge 2022
- Invite-only for the winners of the Robust Vision Challenge 2022 from the ECCV 2022 RVC Workshop at the end of October 2022. See robustvision.net for participation. The top three entries for each computer vision task can submit a paper for Track 2 describing their approach and associated topics from Track 1. While track 1 focuses on innovative methods, the entries in track 2 are allowed to contain more technical implementation details and combinations of existing techniques which lead to the winning entries.
Submitted papers should present original, unpublished work, relevant to one of the topics of the Special Issue. All submitted papers will be evaluated on the basis of relevance, significance of contribution, technical quality, scholarship, and quality of presentation, by at least three independent reviewers. It is the policy of the journal that no submission, or substantially overlapping submission, be published or be under review at another journal or conference at any time during the review process. Manuscripts will be subject to a peer reviewing process and must conform to the author guide lines available on the IJCV website at: https://www.springer.com/11263
For Track 1 please select "S.I.: Track 1: Robust Vision" at the beginning of the submission process.
For Track 2 please select "S.I.: Track 2: Invite only- Robust Vision Challenge 2022
Authors are encouraged to submit high-quality, original work that has neither appeared in, nor is under consideration by other journals. All papers will be reviewed following standard reviewing procedures for the Journal. Papers must be prepared in accordance with the Journal guidelines: www.springer.com/11263
Springer provides a host of information about publishing in a Springer Journal on our Journal Author Resources page, including FAQs, Tutorials along with Help and Support.
Other links include:
For authorsSubmit manuscript
Working on a manuscript?
Avoid the most common mistakes and prepare your manuscript for journal editors.Learn more