search

zhu-xlab / DOFA

Code for Neural Plasticity-Inspired Foundation Model for Observing the Earth Crossing Modalities
MIT License
84 stars 5 forks source link

Question - Augmentation for downstream tasks to recreate GeoBench values from the paper #16

Open sunny1401 opened 1 month ago

sunny1401 commented 1 month ago

Hi - Thank you for this work - it is very useful. I am trying to run DOFA on downstream tasks from scratch, to try and recreate some value in the paper for GeoBench dataset I wanted to make sure I get the data processing correct,

I am using this transformations for train.

Compose([
    RandomResizedCrop((size, size), scale=(0.8, 1)), 
    RandomHorizontalFlip()
    Normalise using relevant band values from geo-bench
])

For Test I just use the above with Resize(size, size) and Normalise.

was wondering if the above is enough to recreate the values or did i miss anything?