sitsfm

Scripts for testing foundational models in SITS

Run in the following order

1. generate_foundation_models.R: reads a set of samples and produces foundational models for the MAE, LeJEPA, VICReg, Barlow Twins, and SUPCON encoders.

2. cerrado_lucc_2024_tiles.R: extract a set of tiles from 2024 LUCC classification.

3. cerrado_emb_class_mae_2024.R, cerrado_emb_class_lejepa_2024.R, cerrado_emb_class_vicreg_2024.R, cerrado_emb_class_btwins_2024.R, cerrado_emb_class_supcon_2024.R: produce embeddings for 2024 LUCC tiles and classify them using samples_v13.

4. cerrado_lucc_2024_urban.R: reclassifies 2024 LUCC tiles based on TerraClass 2024 information on urban areas and water bodies.

5. cerrado_embeddings_benchmark_urban.R: uses the 2024 LUCC tiles to produce a set of stratified samples, calculates accuracy assessment for the classifications produced by each embeddings, and stores the results as a XLSX file.