Roboflow100-VL
收藏资源简介:
Roboflow100-VL 是一个大规模的多模态目标检测数据集,包含 100 个不同领域的多模态对象检测数据集,旨在评估视觉语言模型 (VLM) 在互联网规模预训练中不常见的概念上的性能。该数据集由 Roboflow Universe 平台提供,包含 564 个类别和 164,149 张图像,超过 1.3 百万个标注。Roboflow100-VL 的创建旨在解决 VLM 在新概念上的泛化问题,通过提供包含少量视觉示例和丰富文本描述的标注指令来实现。该数据集适用于零样本、少样本、半监督和全监督学习场景,旨在推动 VLM 在真实世界应用中的发展。
Roboflow100-VL is a large-scale multimodal object detection dataset composed of 100 multimodal object detection datasets across diverse domains. It is designed to evaluate the performance of Vision-Language Models (VLMs) on concepts that are rare in internet-scale pre-training datasets. Hosted on the Roboflow Universe platform, this dataset contains 564 categories, 164,149 images, and over 1.3 million annotations. Developed to address the generalization challenge of VLMs when facing novel concepts, Roboflow100-VL provides annotation instructions with a small number of visual examples and rich textual descriptions. This dataset supports zero-shot, few-shot, semi-supervised, and fully-supervised learning scenarios, and aims to advance the real-world application and development of VLMs.

- 1通过卡内基梅隆大学 · 2025年



