Abstract: Aimed at the problem that weakly supervised action localization method cannot locate action directly and the localization accuracy was not advanced, a weakly supervised action localization method based on the attention mechanism was proposed, and an action localization model based on the pre and post information of action frame and the distinguishing function was designed and realized. Conditional Variational AutoEncoder (CVAE) attention value generation model was used to generate frame-level attention values, which was used as pseudo-frame-level labels. In order to enhance the pre and post relevance of frame, CVAE attention value generation model was improved, to obtain the frame-level attention value, the pre and post information of action frame was added; To train and optimize pseudo-frame-level labels repeatedly, the optimization model for attention value based on distinguishing function was used. The experimental results conducted on THUMOS14 and ActivityNet1.2 datasets show that the action localization model based on the pre and post information of the action frame and the distinguishing function has better action localization effect and accuracy, which makes the movement of missing detection rate reduced by 11.7 percentage compared with the model without the pre and post information of action frame; Compared with AutoLoc, Weakly-supervised Temporal Activity Localization and Classification framework (W-TALC), 3C-Net and other weakly supervised action localization models, when IoU (Intersection over Union) value is setted to 0.5, the mean Average Precision (mAP) value on THUMOS14 dataset is improved by more than 10.7 percentage, and the mAP value on ActivityNet1.2 dataset is improved by more than 8.8 percentage.