地址:https://github.com/LUSSeg/ImageNet-S 1 例图 2. 类别和数量信息 疑问 根据原文的描述:Based on the ImageNet dataset, we propose the ImageNet-S dataset with 1.2 million training images and 50k high-quality semantic segmentation annotations for evaluation。 只标注了5w张测试集??那训练集是怎么来的?