train.md

Train GeoSynth

🧑‍💻 Setting up environment

Create a conda environment:

conda env create -f environment.yaml
conda activate control

💬 Extracting LLaVA Captions

We use LLaVA-7b for automatically captioning each satellite image. Run the following command to get captions corresponding to each satellite image:

python get_llava_captions.py

🛰️ Extracting SatCLIP Embeddings

Execute the following command to get all the geographic location embeddings:

python get_satclip_embeds.py

🔥 Training

Setup all parameters of interest in train.py, then run:

python train.py