Abstract Weakly supervised instance segmentation with imagelevel labels, instead of expensive pixel-level masks, remains unexplored. In this paper, we tackle this challenging problem by exploiting class peak responses to enable a classi- fification network for instance mask extraction. With image labels supervision only, CNN classififiers in a fully convolutional manner can produce class response maps, which specify classifification confifidence at each image location. We observed that local maximums, i.e., peaks, in a class response map typically correspond to strong visual cues residing inside each instance. Motivated by this, we fifirst design a process to stimulate peaks to emerge from a class response map. The emerged peaks are then back-propagated and effectively mapped to highly informative regions of each object instance, such as instance boundaries. We refer to the above maps generated from class peak responses as Peak Response Maps (PRMs). PRMs provide a fifine-detailed instance-level representation, which allows instance masks to be extracted even with some off-the-shelf methods. To the best of our knowledge, we for the fifirst time report results for the challenging image-level supervised instance segmentation task. Extensive experiments show that our method also boosts weakly supervised pointwise localization as well as semantic segmentation performance, and reports state-ofthe-art results on popular benchmarks, including PASCAL VOC 2012 and MS COCO