LAMM
收藏资源简介:
LAMM数据集是由上海人工智能实验室开发的多模态指令调优数据集,旨在通过语言辅助提升多模态大语言模型(MLLMs)在视觉任务上的表现。该数据集包含丰富的视觉任务数据,如2D和3D视觉任务,以及高质量的指令和响应。通过使用GPT-API生成指令和响应,数据集强调了细粒度信息和事实知识,增强了MLLMs对视觉任务指令的理解和泛化能力。LAMM数据集的应用领域广泛,包括图像和点云处理,旨在解决多模态交互中的理解和执行差距,促进人机无缝交互。
The LAMM dataset is a multimodal instruction-tuning dataset developed by the Shanghai AI Laboratory, aiming to enhance the performance of multimodal large language models (MLLMs) on visual tasks via language assistance. This dataset includes rich visual task data, such as 2D and 3D visual tasks, as well as high-quality instruction-response pairs. It generates instructions and responses using the GPT-API, emphasizing fine-grained information and factual knowledge to improve MLLMs' understanding and generalization capabilities for visual task instructions. The LAMM dataset has broad application fields, covering image and point cloud processing, and aims to address the understanding-execution gap in multimodal interactions and promote seamless human-computer interaction.




