DenseLayout is a benchmark for Layout-to-Image (L2I) generation in dense scenes. Each image contains 15+ instances on average with bounding boxes, categories, and captions. The dataset supports evaluation from:
Region level – spatial alignment and attribute accuracy
Global level – overall image quality and prompt faithfulness