ONE-Lab/MultiRef-benchmark
收藏Hugging Face2025-12-14 更新2025-10-18 收录
下载链接：
https://hf-mirror.com/datasets/ONE-Lab/MultiRef-benchmark
下载链接
链接失效反馈官方服务：
资源简介：
MultiRef-Benchmark是一个包含1,990个具有挑战性的样本的评估框架，用于测试可控图像生成模型整合来自多个来源的不同视觉元素的能力。这些样本分为真实世界样本和合成样本，旨在评估模型在遵循复杂视觉指令、整合冲突或互补视觉参考等方面的性能。
MultiRef-Benchmark is an evaluation framework containing 1,990 challenging samples designed to test the ability of controllable image generation models to integrate different visual elements from multiple sources. These samples are split into real-world and synthetic, aiming to evaluate the models performance in following complex visual instructions and integrating conflicting or complementary visual references.
提供机构：
ONE-Lab
