IMPROVING SCENE GRAPH GENERATION WITH LLMS

Summary from the TLDR AI newsletter

Weakly-Supervised Scene Graph Generation (WSSGG) using image captions often faces challenges in capturing detailed relationships and utilizing all available information. This study introduces LLM4SGG, which leverages the capabilities of large language models to better extract and align relationships from captions.