ImageNet statistics and PCA

ImageNet-1k's covariance matrix's eigenvalues (eigenvalues_ipca.npy), the ratio of total variance explained by each of ImageNet-1k's principal component (eigenvalues_ratio_ipca.npy), ImageNet-1k's principal components (pc_matrix_ipca.npy) computed using the normalized training dataset. For computational reasons, only 10% of the training dataset was used for PCA and only the top 20k principal components were computed.

These items were used in [1]. The ImageNet-1k dataset was presented in [2].

[1] Alice Bizeul, Thomas M. Sutter, Alain Ryser, Julius Von Kügelgen, Bernhard Schölkopf, Julia E. Vogt. Components Beat Patches: Eigenvector Masking for Visual Representation Learning. Oct, 2024.

[2] Deng, Jia, et al. "Imagenet: A large-scale hierarchical image database." 2009 IEEE conference on computer vision and pattern recognition. Ieee, 2009.