MAviS: A Multimodal Conversational Assistant For Avian Species
Researchers have developed MAviS, a multimodal conversational AI designed for understanding avian species. This system utilizes a new dataset, MAviS-Dataset, which combines image, audio, and text data for over 1,000 bird species. MAviS-Chat, the model built on this dataset, demonstrates superior performance in species-specific question answering and scene description compared to existing models. A benchmark, MAviS-Bench, was also created to evaluate these capabilities. AI
IMPACT Domain-specific multimodal LLMs can improve ecological monitoring and biodiversity conservation efforts.