DABS
收藏资源简介:
DABS是一个跨领域的自监督学习基准数据集,由斯坦福大学创建,包含七个不同领域的数据集:自然图像、多通道传感器数据、英语文本、语音记录、多语言文本、胸部X光片和带有文本描述的图像。每个领域都包含一个用于预训练的无标签数据集,以及用于评估模型下游性能的标记任务集。DABS旨在推动领域无关的自监督学习算法的发展,解决不同领域间的数据标注难题,并探索自监督学习在多模态环境中的应用。
DABS is a cross-domain self-supervised learning benchmark dataset developed by Stanford University. It comprises seven distinct domains: natural images, multi-channel sensor data, English text, speech recordings, multilingual text, chest X-rays, and images with textual descriptions. Each domain includes an unlabeled dataset for pre-training, as well as a labeled task set for evaluating the downstream performance of models. DABS aims to advance the development of domain-agnostic self-supervised learning algorithms, address the challenge of data annotation across diverse domains, and explore the application of self-supervised learning in multimodal environments.

- 1DABS: A Domain-Agnostic Benchmark for Self-Supervised Learning斯坦福大学 · 2023年



