0xLDF/SACap-1M
收藏Hugging Face2025-08-11 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/0xLDF/SACap-1M
下载链接
链接失效反馈官方服务：
资源简介：
SACap-1M是一个大规模的开放式词汇分割掩膜到图像生成数据集，来源于高分辨率的SA-1B数据集。它包含1百万张图像和5.9百万个实例级分割掩膜。每个掩膜都带有由Qwen2-VL-72B生成的平均14.1个单词的区域标题，每张图像都配有一个平均58.6个单词的全局标题。
SACap-1M is a large-scale, open-vocabulary dataset for segmentation-mask-to-image generation, sourced from the high-resolution SA-1B. It contains 1 million images and 5.9 million instance-level segmentation masks. Each mask is annotated with a regional caption (average 14.1 words) generated by Qwen2-VL-72B, and every image is paired with a global caption (average 58.6 words).
提供机构：
0xLDF
