Weakly supervised temporal action localization is a challenging vision task\ndue to the absence of ground-truth temporal locations of actions in the\ntraining videos. With only video-level supervision during training, most\nexisting methods rely on a Multiple Instance Learning (MIL) framework to\npredict the start and end frame of each action category in a video. However,\nthe existing MIL-based approach has a major limitation of only capturing the\nmost discriminative frames of an action, ignoring the full extent of an\nactivity. Moreover, these methods cannot model background activity effectively,\nwhich plays an important role in localizing foreground activities. In this\npaper, we present a novel framework named HAM-Net with a hybrid attention\nmechanism which includes temporal soft, semi-soft and hard attentions to\naddress these issues. Our temporal soft attention module, guided by an\nauxiliary background class in the classification module, models the background\nactivity by introducing an "action-ness" score for each video snippet.\nMoreover, our temporal semi-soft and hard attention modules, calculating two\nattention scores for each video snippet, help to focus on the less\ndiscriminative frames of an action to capture the full action boundary. Our\nproposed approach outperforms recent state-of-the-art methods by at least 2.2%\nmAP at IoU threshold 0.5 on the THUMOS14 dataset, and by at least 1.3% mAP at\nIoU threshold 0.75 on the ActivityNet1.2 dataset. Code can be found at:\nhttps://github.com/asrafulashiq/hamnet.\n