AWCV-100K

Mainstream real-world devices typically utilize standard RGB cameras and necessitate comprehensive coverage of commonly used words for effective communication. To tackle these challenges, we introduce AWCV-100K, a pioneering video-based air-writing dataset designed for real-world applications. AWCV-100K is a video-based air-writing Chinese character dataset, including 102,688 videos and comprising a total of 8.8 million video frames. The benchmark includes 3,755 Chinese characters from the GB1 set, encompassing 99.7% of characters used in daily communication, forming a comprehensive corpus. To ensure the developed model is robust to variations among different individuals, we have partitioned the dataset into three sets (i.e., training, validation, and testing) in an approximate ratio of 8:1:1, dividing the data by person.主流实际应用场景中的设备通常搭载标准RGB相机，且需覆盖常用汉字以实现高效的空中书写交互。为解决此类应用挑战，我们推出AWCV-100K——一款专为实际场景打造的开创性视频类空中书写数据集。AWCV-100K是基于视频的空中书写汉字数据集，包含102688段视频，总视频帧数量达880万。该基准数据集涵盖GB1字符集中的3755个汉字，覆盖日常通信所用汉字的99.7%，构成了一套覆盖全面的语料库。为确保所开发模型对不同个体的书写差异具备鲁棒性，我们按个体维度对数据集进行划分，将其分为训练集、验证集与测试集三个子集，划分比例约为8:1:1。