Dense-World/Sa2VA-Training
收藏Hugging Face2025-09-08 更新2025-09-13 收录
下载链接：
https://hf-mirror.com/datasets/Dense-World/Sa2VA-Training
下载链接
链接失效反馈官方服务：
资源简介：
Sa2VA数据集是一个用于图像和文本到文本任务的数据集，支持多种图像和视频任务，如引用分割和对话。该数据集通过结合SAM-2视频分割模型和LLaVA视觉语言模型，将文本、图像和视频信息统一到共享的LLM标记空间中，实现了对图像和视频的密集地面理解。
The Sa2VA dataset is designed for image-text-to-text tasks, supporting a variety of image and video tasks such as referring segmentation and conversation. By integrating the SAM-2 video segmentation model with the LLaVA vision-language model, the dataset unifies text, image, and video information into a shared LLM token space for dense grounded understanding of images and videos.
提供机构：
Dense-World
