dopanim
收藏资源简介:
dopanim数据集由卡塞尔大学创建,包含约15,750张动物图像,涵盖15个类别,特别关注分类难度较高的动物类别。数据集通过iNaturalist平台收集,确保了图像的高质量和多样性。数据集的创建过程中,20名标注者提供了超过52,000个标注,准确率约为67%。该数据集主要用于研究多标注者学习、噪声标签学习和主动学习等领域,旨在提高机器学习模型对噪声标注的鲁棒性。
The Dopanim dataset was created by Kassel University. It consists of approximately 15,750 animal images spanning 15 categories, with a particular focus on animal classes that are relatively difficult to classify. The dataset was collected via the iNaturalist platform, ensuring high image quality and diversity. During its development, 20 annotators provided over 52,000 annotations, with an overall annotation accuracy of approximately 67%. This dataset is primarily used for research in multi-annotator learning, noisy label learning, active learning and other related fields, aiming to improve the robustness of machine learning models against noisy annotations.

- 1dopanim: A Dataset of Doppelganger Animals with Noisy Annotations from Multiple Humans卡塞尔大学 · 2024年



