DSText V2
收藏资源简介:
DSText V2是由浙江大学创建的综合性视频文本检测数据集,专注于密集和小型文本的挑战。该数据集包含140个视频片段,总计62.1k视频帧和2.2m文本实例,来源于YouTube,覆盖7个开放场景。数据集的创建过程涉及从BOVText、RoadText-1k和YouTube收集视频,并由专业团队进行标注。DSText V2支持视频文本检测、跟踪和端到端视频文本识别三个任务,旨在解决视频文本识别中的密集和小型文本挑战,为计算机视觉社区提供新的研究视角和基准。
DSText V2 is a comprehensive video text detection dataset developed by Zhejiang University, focusing on the challenges of dense and small-scale text. This dataset contains 140 video clips, totaling 62.1k video frames and 2.2 million text instances, sourced from YouTube and covering 7 open scenarios. The dataset creation process involves collecting videos from BOVText, RoadText-1k and YouTube, followed by annotation by a professional team. DSText V2 supports three tasks, namely video text detection, tracking and end-to-end video text recognition, aiming to address the challenges of dense and small-scale text in video text recognition, and provide new research perspectives and benchmarks for the computer vision community.




