ImageNet-V2

ImageNet-V2 数据集包含 ImageNet 基准测试的新测试数据。 ImageNet-V2 包含三个测试集，每个测试集包含 10,000 张新图像。重要的是，这些测试集是在原始 ImageNet 数据集经过十年的进展后采样的。这使得新的测试数据独立于现有模型，并保证准确性分数不受自适应过度拟合的影响。

The ImageNet-V2 dataset contains new test data for the ImageNet benchmark. It comprises three test sets, each holding 10,000 newly curated images. Critically, these test sets were sampled over a decade after the original ImageNet dataset was released. This ensures that the new test data is independent of existing models, and guarantees that accuracy scores are not impacted by adaptive overfitting.

搜集汇总

背景与挑战

背景概述

ImageNet-V2是ImageNet基准测试的新测试数据集，包含三个测试集，每个测试集约10,000张图像。这些图像在原始数据集发布十年后采样，旨在独立评估模型性能，避免自适应过度拟合的影响。