LayoutSAM-Eval is a comprehensive benchmark for evaluating the quality of Layout-to-Image (L2I) generation models. This benchmark assesses L2I generation quality from two perspectives: region-wise quality (spatial and attribute accuracy) and global-wise quality (visual quality and prompt following). It employs the VLM’s visual question answering to evaluate spatial and attribute adherence, and utilizes various metrics including IR score… See the full description on the dataset page: https://huggingface.co/datasets/HuiZhang0812/LayoutSAM-eval.
