tieredImageNet
收藏资源简介:
tieredImageNet 数据集是 ILSVRC-12 的一个更大的子集,在 ImageNet 人工策划的层次结构中,有 608 个类(779,165 个图像)被分组到 34 个更高级别的节点中。这组节点被划分为 20、6 和 8 个不相交的训练、验证和测试节点集,相应的类形成各自的元集。正如 Ren 等人所争论的那样。 (2018 年),ImageNet 层次结构根附近的这种分裂导致了一个更具挑战性但更现实的制度,其中测试类与训练类不太相似。
The tieredImageNet dataset is a larger subset of ILSVRC-12. Within the manually curated ImageNet hierarchy, 608 classes (comprising a total of 779,165 images) are grouped into 34 higher-level nodes. These nodes are split into three disjoint sets with sizes 20, 6, and 8 for training, validation, and testing respectively, and the corresponding classes form their respective meta-sets. As argued by Ren et al. (2018), this split near the root of the ImageNet hierarchy yields a more challenging yet more realistic experimental setting, in which test classes exhibit low similarity to training classes.




