Zero-Shot Land Cover Classification of SAR-like Imagery using Vision-Language Models

This capsule contains the reproducibility code for the IEEE GRSL submission "Zero-Shot Land Cover Classification of SAR-like Imagery using Vision-Language Models". It implements a comprehensive zero-shot classification framework using CLIP on the EuroSAT dataset. The code includes pipelines for: (1) converting optical imagery to SAR-like intensity, (2) evaluating 10+ prompt engineering strategies strategies (comparing spatial vs. domain-specific prompts), and (3) performing hierarchical classification (Coarse vs. Fine).