Occluded Video Instance Segmentation with Set Prediction Approach

Occluded Video Instance Segmentation (OVIS) is a multi-task problem performing detection, segmentation, and tracking simultaneously under severe occlusions. We propose an extended model for the OVIS task based on the real-time one-stage instance segmentation method. The proposed model was applied to the OVIS dataset hold by the ICCV 2021 - Occluded Video Instance Segmentation Workshop 2021. We also show that the occlusions can be handled efficiently through one-stage approaches.