International Workshop on Learning from Small Data in Computer Vision (SmallData-CV 2026)
A workshop associated with the Asian Conference on Computer Vision (ACCV 2026)
Osaka, Japan | Monday, 14 December 2026 (Half-day format, flexible to full-day)
International Workshop on Learning from Small Data in Computer Vision (SmallData-CV 2026)
A workshop associated with the Asian Conference on Computer Vision (ACCV 2026)
Osaka, Japan | Monday, 14 December 2026 (Half-day format, flexible to full-day)
Aim and Scope
Learning from small datasets remains a fundamental bottleneck in computer vision, limiting the deployment of reliable AI in high-stakes and resource-constrained domains. While large-scale benchmarks have driven impressive advances, many real-world applications—ranging from medical imaging and robotics to autonomous driving and remote sensing—cannot acquire large, richly annotated image collections due to high annotation costs, privacy regulations, data scarcity, or the inherent rarity of target events. This challenge has spurred a diverse set of strategies, including data augmentation, synthetic generation using GANs and diffusion models, transfer learning, few-shot and meta-learning, semi-supervised and self-supervised learning, active learning, in-context adaptation, and the incorporation of prior knowledge such as 3D geometry and physical laws. However, deploying these methods in practice raises critical concerns about robustness, generalization, and reliability. Models that succeed on controlled benchmarks often falter under domain shift, changing lighting and camera angles, imbalanced classes, data drift, and noisy or uncertain labels. Moreover, in sensitive application areas, learning must respect privacy constraints, handle streaming high-dimensional data, and operate efficiently in dynamic environments. This workshop provides a dedicated forum for researchers and practitioners to explore the frontiers of small-data learning in computer vision. We aim to bridge algorithmic innovation with practical deployment, bringing together experts in machine learning, computer vision, and applied domains to address both theoretical advances and the real-world barriers to data-efficient vision systems.
Call for Papers
The reliance on massive labeled datasets remains a critical obstacle in computer vision. While large-scale benchmarks have fueled remarkable progress, numerous high-impact applications—from autonomous navigation and medical diagnostics to industrial inspection, robotics, and environmental monitoring—cannot access large, fully-annotated image collections due to prohibitive annotation costs, privacy regulations, equipment constraints, or the inherent scarcity of target events. This challenge has motivated a surge of research into data-efficient paradigms, including data augmentation, synthetic generation, transfer learning, few-shot and meta-learning, active learning, semi-supervised and self-supervised learning, in-context adaptation, and knowledge-guided approaches. However, translating these algorithmic advances into robust, real-world systems demands careful attention to generalization under domain shift, resilience to label noise, reliability in dynamic environments, and efficient handling of streaming high-dimensional data.
The SmallData-CV 2026 Workshop invites original contributions that advance the theory, algorithms, and practical deployment of learning methodologies for data-scarce vision problems. We aim to unite researchers, engineers, and domain experts to explore not only algorithmic performance, but also the robustness, efficiency, interpretability, and adaptability of models trained with limited supervision. The workshop provides a forum for bridging cutting-edge machine learning with real-world constraints, fostering interdisciplinary exchange between computer vision, applied mathematics, and domain-specific fields.
We welcome methodological, empirical, applied, and position papers addressing small-data learning across all vision domains, including but not limited to robotic vision, medical imaging, autonomous driving, remote sensing, augmented/virtual reality, and industrial quality control. Contributions involving novel learning frameworks, benchmark studies, or critical analyses of existing approaches are equally encouraged.
Topics of interest include, but are not limited to:
Efficient fine-tuning and adaptation of pre-trained and foundation models for vision
Data augmentation, synthetic image generation, and physics-based rendering (GANs, diffusion models, 3D simulation)
Few-shot, zero-shot, and meta-learning for image and video analysis
Active learning and optimal query strategies for efficient annotation
Semi-supervised and self-supervised representation learning with limited labels
In-context learning and dynamic model adaptation
Knowledge-guided vision (incorporating 3D geometry, physical laws, and scene constraints)
Handling imbalanced classes, noisy labels, and label scarcity
Semi-supervised and federated learning for privacy-preserving vision
Uncertainty quantification, calibration, and reliability estimation
Adaptive streaming and online learning for changing visual environments
Surrogate models and Bayesian optimization for efficient experiment design
We particularly encourage submissions that move beyond conventional benchmark performance and provide rigorous evidence of generalization across varying camera viewpoints, lighting conditions, sensor modalities, and deployment environments. Studies involving real-world validation, multi-scenario evaluation, failure analysis, or integration with physical systems are highly welcomed. Position papers that critically assess the current state, identify open challenges, or propose new research directions in small-data computer vision are also strongly encouraged.
All submissions will undergo a rigorous peer-review process. Accepted contributions will be presented during the workshop as oral or poster presentations, offering a valuable opportunity for scientific exchange and interdisciplinary discussion. Detailed submission instructions, formatting requirements, and important deadlines are provided below.
Alaa Tharwat – Bielefeld University of Applied Sciences and Arts (HSBI), Germany
Essam Rashed – University of Hyogo, Japan
Ghada Khoriba – Nile University, Egypt
Wolfram Schenck – Bielefeld University of Applied Sciences and Arts (HSBI), Germany
TBA
Paper Submission Timeline
The workshop invites original research, short, and position papers, with full papers limited to 6–8 pages and short papers to 4 pages (excluding references). All submissions will be evaluated through a double-blind peer-review process involving at least two reviewers. Papers will be judged on technical quality, novelty, relevance, clarity, reproducibility, and their potential contribution to trustworthy medical foundation models.
CFP Released: 5 August 2026
Submission Deadline: 25 September 2026
Author Notification: 13 October 2026
Camera-Ready Deadline: 20 October 2026
Workshop Date: 14 December (PM) 2026 (Room: 1001)
Submission link: https://openreview.net/group?id=afcv.org/ACCV/2026/Workshop/LFSD-CV
Awards and Publication Opportunities
Following the ACCV 2026 conference rules, best papers by students and researchers will be recognized, and the winning papers will be invited for publication in a special issue of the International Journal of Computer Vision (IJCV).
In addition, the journal Machine Learning and Knowledge Extraction (MAKE) offers APC discounts for the top three papers from our workshop:
1st place: Full APC waiver + 200 CHF
2nd place: 50% discount + 150 CHF
3rd place: 30% discount + 100 CHF
Program
This workshop will be held on Monday 14 December, 2026, in-person and spans a half-day (or full-day depends on the number of submissions). The program will feature peer-reviewed oral presentations, a dedicated interactive poster session, and a concluding panel discussion (tentative).
The tentative schedule is as follows:
13:00 – 13:15 Opening Remarks
13:15 – 15:00: Oral Session 1
15:00 – 15:30: Coffee Break & Poster Session
15:30 – 17:30: Oral Session 2
17:30 – 18:00: Panel Discussion + Closing Remarks + Best papers announcement
Venue: Osaka International Convention Center (Grand Cube Osaka), Room 1001
Contact
For any questions, please contact the primary organizer:
Alaa Tharwat – alaa.othman@hsbi.de