This research introduces a novel approach to zero-shot fungal classification by combining synthetic data generation with image captioning techniques.

• Leverages large language models to generate descriptive text for various fungal growth stages
• Creates synthetic fungi image datasets specifically designed to improve classification accuracy
• Employs vision-language models like CLIP to enable identification of previously unseen fungi species
• Demonstrates how artificial data sources can overcome limitations in specialized biological datasets

This work offers significant value for biological sciences by expanding automated identification capabilities without requiring extensive labeled datasets for every fungal species, potentially accelerating research and fieldwork in mycology.

FungalZSL: Zero-Shot Fungal Classification with Image Captioning Using a Synthetic Data Approach