How to get the WILDS RxRx1 dataset with 6 (rather than 3) spectral channels?

[LEMTideman]

Dear WILDS team,

I am looking for the code that you used to process the original RxRx1 dataset into the RxRx1-WILDS dataset. By "original" dataset, I mean the dataset shared by Recursion (https://www.rxrx.ai/) for the CellSignal NeurIPS 2019 competition hosted on Kaggle (https://www.kaggle.com/c/recursion-cellular-image-classification). I prefer the RxRx1-WILDS dataset because the WILDS dataset class is well structured, and I like that the images in RxRx1-WILDS are smaller than the original images (256 × 256 centre crops, rather than 512 × 512 centre crops of the downsampled images). However, the RxRx1-WILDS images only have 3 spectral channels. I want to use all 6 channels of the autofluorescence microscopy data (nuclei, endoplasmic reticuli, actin, nucleoli and cytoplasmic RNA, mitochondria, and Golgi). What code did you use to process the original RxRx1 dataset into the RxRx1-WILDS dataset?

Thanks!