Viewpoint constrained and unconstrained Cricket stroke localization from untrimmed videos
作者：
Highlights：
• We create two Cricket strokes datasets, with start and end frame annotations of strokes played in untrimmed videos.
• Solve the temporal Cricket stroke localization problem by building viewpoint Constrained and Unconstrained pipelines.
• Two novel post-processing ideas, of longest subsequence selection and boundary correction, are also described.
• We compared our results with Segment CNNs and TRNs.
• The best result on Generic test set partition was obtained by using our proposed constrained model.
摘要
•We create two Cricket strokes datasets, with start and end frame annotations of strokes played in untrimmed videos.•Solve the temporal Cricket stroke localization problem by building viewpoint Constrained and Unconstrained pipelines.•Two novel post-processing ideas, of longest subsequence selection and boundary correction, are also described.•We compared our results with Segment CNNs and TRNs.•The best result on Generic test set partition was obtained by using our proposed constrained model.
论文关键词：Cricket stroke,Action localization,TIoU,C3D,RNN,GRU
论文评审过程：Received 14 May 2020, Accepted 16 May 2020, Available online 29 May 2020, Version of Record 17 June 2020.
论文官网地址：https://doi.org/10.1016/j.imavis.2020.103944
