One ICCV-2023 paper to appear: SeiT: Storage-Efficient Vision Training with Tokens Using 1% of Pixel Storage.