BenchLMM
收藏资源简介:
BenchLMM是由南洋理工大学开发的用于评估大型多模态模型在跨风格视觉能力上的基准数据集。该数据集包含500张图像,涵盖艺术图像风格、成像传感器风格和应用风格三大类,每类下又细分为五个子风格。数据集旨在解决现有模型在处理不同风格图像时的性能下降问题,通过提供多样化的视觉分布变化,帮助开发更智能和多才多艺的大型多模态模型。
BenchLMM is a benchmark dataset developed by Nanyang Technological University for evaluating the cross-style visual capabilities of large multimodal models. This dataset consists of 500 images, covering three major categories: artistic image styles, imaging sensor styles, and application styles, with five sub-styles under each category. It aims to address the performance degradation issue of existing models when processing images of different styles, and assists in developing more intelligent and versatile large multimodal models by providing diverse variations in visual distributions.




