IRT Datasets
收藏NIAID Data Ecosystem2026-05-01 收录
下载链接：
https://zenodo.org/record/7966685
下载链接
链接失效反馈官方服务：
资源简介：
New Test Sets of the Paper "Are All Test Samples Equal? An Item Response
Theory-based Study in Computer Vision".
Test Set 1(As presented in the main paper)
TOP50K_1PL_IRT: Sampled the top 50,000 images sorted in descending order of their difficulty scores.
TOP50K_Uniform_Random: Sampled 50,000 images at random.
Test Set 2(As presented in the supplementary paper)
Class_wise_1PL_IRT: Sampled the top 50 most difficult images from each of the 1000 classes, resulting in a test set of 50,000 images.
Class_wise_Uniform_Random: Sampled 50 images at random from each of the 1000 classes.
创建时间：
2023-05-25
