Normalizing Channels for PovertyMap

[niki-amini-naieni]

Hello,

In the Wilds codebase, I do not see any normalization of the images loaded from the PovertyMap dataset. However, in appendix E.8.4, in the data processing section, the corresponding paper reads "We normalize each channel by the pixel-wise mean and standard deviation for each channel." Has this normalization already been done before loading the PovertyMap data with the provided data loaders in the WILDS codebase?

Link to original paper: https://arxiv.org/pdf/2012.07421.pdf

Thanks!

[kohpangwei]

Yup, it's been pre-normalized. See here: https://github.com/p-lambda/wilds/blob/main/wilds/datasets/poverty_dataset.py#L69

and the corresponding dataset processing function that passes in normalize=True: https://github.com/p-lambda/wilds/blob/main/dataset_preprocessing/poverty/batcher.py#L63

[niki-amini-naieni]

Great, thank you!