YOLACT is a real-time instance segmentation model that achieves 29.8 mAP on the MS COCO dataset at 33.5 fps, significantly outperforming previous competitive approaches. The model breaks the instance segmentation task into two parallel subtasks: generating a set of prototype masks and predicting per-instance mask coefficients. These components are then combined to produce high-quality instance masks. YOLACT's approach is fully convolutional and does not require explicit feature localization, making it faster and more efficient. The model learns to localize instances on its own, producing spatially coherent masks without the need for re-pooling. YOLACT also introduces Fast NMS, a 12 ms faster replacement for standard NMS with minimal performance penalty. The code for YOLACT is available at \url{https://github.com/dbolya/yolact}.YOLACT is a real-time instance segmentation model that achieves 29.8 mAP on the MS COCO dataset at 33.5 fps, significantly outperforming previous competitive approaches. The model breaks the instance segmentation task into two parallel subtasks: generating a set of prototype masks and predicting per-instance mask coefficients. These components are then combined to produce high-quality instance masks. YOLACT's approach is fully convolutional and does not require explicit feature localization, making it faster and more efficient. The model learns to localize instances on its own, producing spatially coherent masks without the need for re-pooling. YOLACT also introduces Fast NMS, a 12 ms faster replacement for standard NMS with minimal performance penalty. The code for YOLACT is available at \url{https://github.com/dbolya/yolact}.