Daksh12081/kenya-contrastive-learning

# Kenya Forest Classification using CROMA ## Overview This repository contains the preprocessing and downstream training pipeline for land cover classification using the pretrained CROMA foundation model on the SEN12MS dataset. ## Dataset - Dataset: SEN12MS Spring - Total Samples: 38,755 - Modalities: - Sentinel-1 SAR - Sentinel-2 Optical ## Classes | ID | Class | |----|----------------| |0|Forest| |1|Shrub| |2|Crop| |3|Grass| |4|NonVegetation| ## Pipeline SEN12MS TIFFs ↓ Dataset Loader ↓ Training Tensors ↓ CROMA Encoder ↓ Embeddings ↓ Multiclass Classifier ## Repository Structure ``` datasets/ scripts/ embeddings/ downstream/ configs/ ``` ## Requirements ``` pip install -r requirements.txt ``` ## Current Status - Dataset preprocessing ✅ - Tensor generation ✅ - GCS upload ✅ - Downstream classifier ✅ - Embedding generation ⏳