SOOD-ImageNet
收藏资源简介:
SOOD-ImageNet是由帕多瓦大学信息工程系创建的一个大规模数据集,旨在解决计算机视觉中的语义分布外(OOD)检测问题。该数据集包含约160万张图像,涵盖56个类别,适用于图像分类和语义分割任务。数据集的创建过程结合了现代视觉-语言模型的自动标注和人工校验,确保了数据的高质量和大规模。SOOD-ImageNet特别关注语义偏移问题,旨在评估模型在面对语义变化时的泛化能力,适用于自动驾驶、农业和废物管理等多个领域。
SOOD-ImageNet is a large-scale dataset developed by the Department of Information Engineering at the University of Padua, targeting the semantic out-of-distribution (OOD) detection task in computer vision. This dataset comprises approximately 1.6 million images across 56 categories, supporting both image classification and semantic segmentation tasks. The construction of SOOD-ImageNet integrates automatic annotation using modern vision-language models and manual verification, ensuring the high quality and large-scale reliability of the dataset. Specifically, SOOD-ImageNet focuses on the semantic shift problem, aiming to evaluate a model's generalization capability when encountering semantic variations, and is applicable to diverse domains including autonomous driving, agriculture, and waste management.

- 1SOOD-ImageNet: a Large-Scale Dataset for Semantic Out-Of-Distribution Image Classification and Semantic Segmentation帕多瓦大学信息工程系 · 2024年



