Complex Dataset Distillation (Comp-DD)
收藏资源简介:
Complex Dataset Distillation (Comp-DD)是由新加坡国立大学和卡内基梅隆大学合作创建的数据集,旨在解决复杂场景下的数据集蒸馏问题。该数据集包含从ImageNet-1K中挑选的十六个子集,分为八个简单和八个复杂子集。数据集的大小和复杂性通过Grad-CAM激活图的高激活区域比例来衡量。创建过程中,研究团队通过Grad-CAM激活图来增强合成图像中的关键判别区域。Comp-DD数据集主要应用于图像分类和数据集蒸馏领域,旨在提高复杂场景下数据集蒸馏的性能。
Complex Dataset Distillation (Comp-DD) is a dataset jointly created by the National University of Singapore and Carnegie Mellon University, aiming to solve the dataset distillation problem in complex scenarios. This dataset comprises 16 subsets selected from ImageNet-1K, which are categorized into 8 simple subsets and 8 complex subsets. The scale and complexity of the dataset are quantified by the proportion of highly activated regions in Grad-CAM activation maps. During the creation process, the research team leveraged Grad-CAM activation maps to enhance the key discriminative regions in synthetic images. The Comp-DD dataset is mainly applied in the fields of image classification and dataset distillation, with the goal of improving the performance of dataset distillation in complex scenarios.




