Abstract
Illegal wildlife trade is being increasingly conducted through online channels, posing a significant risk to global biodiversity and environmental sustainability. In this paper, we propose a method that employs Vision-Language Models (VLMs) and Large Language Models (LLMs) to analyze online advertisements for wildlife products and generate a description that includes the product type, species, and its status on the IUCN red list and CITES appendices. This detailed information aims to help law enforcement and conservationists in identifying the conservation and trade status, which can be used to tag or filter irrelevant listings and aid in combating illegal wildlife trade. Moreover, we apply our methods on a case study focusing on shark products being sold online. Sharks constitute one of the most heavily traded species, and shark-derived products, such as jaws and vertebrae from threatened shark species, form a substantial part of this trade, making them an urgent conservation matter. Through this study, we are able to demonstrate the effectiveness of our method and its adaptability on niches of wildlife trade by solving an additional problem of classifying products based on the body part of the shark, an out-of-distribution task. This shows the ability to handle problems beyond the original scope. Our flexible approach incorporates modularity of individual models, which allows for improvement when newer models are available, making it a valuable tool for conservationists and law enforcement to curb the threat of illegal online wildlife trade.