In the field of spatio-temporal action detection,some current studies try to solve the spatio-temporal action detec-tion problem using one-stage action detector based on anchor-free.The paper proposes CFAMMOC,a single-stage detector with-out anchor,which is built on the basis of the detector MOC,and two improvements are made in this paper:1)to effectively extract spatio-temporal context information,a channel fusion and attention module is explored to enhance the feature discriminability by ag-gregating the features of the extracted single frames.2)An IOU loss function is added to the original loss function to enhance the re-gression capability of the model on the bounding box of action instances.It conducts experiments on two benchmark datasets,JHM-DB and UCF101-24,and the detection performance is achieved comparable to several state of the art methods.
关键词
时空动作检测/自注意力/IOU损失函数/深度学习
Key words
spatio-temporal action detection/self attention/IOU loss function/deep learning
维普
万方数据