Multi-modal inputs for a given ground sampling distance

[lzachmann]

I noticed in the regression example a list of wavelengths that appears to have included bands from Sentinel-1, alongside the Sentinel-2 bands: https://github.com/Clay-foundation/model/blob/main/finetune/regression/biomasters_model.py#L67-L82. Let's say we were to add that back in.... Is Clay able to generate a single set of n=768 embeddings for a combined set of Sentinel-1 + Sentinel-2 inputs (on the assumption we create a new metadata entry that includes bands from both instruments)? Or is it preferred to compute the embeddings for each sensor separately?

[srmsoumya]

We can pass the combined chip of Sentinel-2 & Sentinel-1 as input to the model. I did try that with the Biomasters example but the loss curves were not very smooth, so I skipped Sentinel 1.