QPredSGG: Hybrid Quantum Predicate Learning for Long-Tailed Scene Graph Generation
Researchers have developed a hybrid quantum-classical model, QPredSGG, to improve scene graph generation by addressing the long-tail imbalance of predicate relations. This model replaces the classical predicate head in the CFEN architecture with a quantum predicate head (QP-Head). The QP-Head significantly reduces the number of trainable parameters while achieving a higher mean recall (mR@100) compared to its classical counterpart. AI
IMPACT Introduces a parameter-efficient quantum approach for complex visual reasoning tasks, potentially improving performance on underrepresented data.