RFC
Project description
NDVI predictions from Sentinel2 Images
Proposed solutions
Using Deep Learning to provide predictions, using meteo, geo, etc..
Implementation choices
PyTorch, maybe Pytorch Lightning, resnet style, point by point, starting from march and predicting onwards.
Constraints
- cluster cloud storage 4 TB
- work on the cluster
- seasonal data planning and availability
- multiple GPUs available
- LOW RAM ON GPU nodes (12GB/gpu), 64GB RAM on CPU nodes
Timeline scheduling
- train the first model before the end of July
- provide first predictions for the wineries before the end of the year