Foundational Few-Shot Object Detection Challenge at CVPR 2026

Roboflow and Carnegie Mellon University have launched the third iteration of the Foundational Few-Shot Object Detection Challenge at CVPR 2026, introducing the Roboflow-20VL dataset to assess foundation models' ability to localize objects from limited visual and textual examples. This new dataset includes 20 diverse domains, such as supermarket product localization and defect detection, with imaging modalities like X-rays and thermal images. The challenge aims to encourage the development of algorithms that can adapt to new domains with minimal examples, addressing the limitations of current models in identifying rare classes. Two competition tracks are available: the Overall Track, allowing any pre-trained model and fine-tuning strategies, and the In-Context Prompting Track, which prohibits gradient-based fine-tuning. Participants are incentivized with cash prizes and are required to submit a technical report, open-source their code, and beat official baselines to qualify. The competition runs from February 20 to May 31, 2026, with top teams recognized at the Workshop On Open World Vision at CVPR 2026.