Video Instance Segmentation (VIS) is a multi-task problem performing\ndetection, segmentation, and tracking simultaneously. Extended from image set\napplications, video data additionally induces the temporal information, which,\nif handled appropriately, is very useful to identify and predict object\nmotions. In this work, we design a unified model to mutually learn these tasks.\nSpecifically, we propose two modules, named Temporally Correlated Instance\nSegmentation (TCIS) and Bidirectional Tracking (BiTrack), to take the benefit\nof the temporal correlation between the object's instance masks across adjacent\nframes. On the other hand, video data is often redundant due to the frame's\noverlap. Our analysis shows that this problem is particularly severe for the\nYoutubeVOS-VIS2021 data. Therefore, we propose a Multi-Source Data (MSD)\ntraining mechanism to compensate for the data deficiency. By combining these\ntechniques with a bag of tricks, the network performance is significantly\nboosted compared to the baseline, and outperforms other methods by a\nconsiderable margin on the YoutubeVOS-VIS 2019 and 2021 datasets.\n