Soo Won Seo

By explicitly modeling relationships between detected instances and their surrounding scene context using VLMs, InCoM-Net achieves state-of-the-art performance in HOI detection, suggesting that holistic scene understanding is crucial for accurate interaction recognition.