SAMText-9M
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/ViTAE-Transformer/SAMText
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个大规模的数据集，包含了超过2400个视频片段，并为视频文本检测生成了超过900万的遮罩标注。该数据集采用了一种遮罩标注流程，这一流程改进了现有的边界框标注方法，能够更好地处理复杂的场景，如密集或弯曲的文本。这些数据来源于2400多个视频片段，共产生了超过900万的遮罩标注，其任务旨在进行视频文本的检测与分割。
This is a large-scale dataset containing over 2400 video clips, with more than 9 million mask annotations generated for video text detection. The dataset adopts a mask annotation workflow that improves upon existing bounding box annotation methods, enabling better handling of complex scenarios such as dense or curved text. With all annotation data derived from these over 2400 video clips, the dataset ultimately yields over 9 million mask annotations, with its target task focused on video text detection and segmentation.
提供机构：
ViTAE-Transformer
