Unbiased Scene Graph Generation

Unbiased Scene Graph Generation (Unbiased SGG) is a crucial task in the field of computer vision, aiming to predict richer scene graphs that include more "tail predicates" by processing skewed and long-tailed distributions of predicate categories. This reduces the model's reliance on high-frequency "head predicates," thereby enhancing the diversity and accuracy of scene understanding. The task holds significant application value for improving fine-grained image content parsing and the recognition of complex relationships.