While Video Instance Segmentation (VIS) has seen rapid progress, current approaches struggle to predict high-quality masks with accurate boundary details. Moreover, the predicted segmentations often fluctuate over time, suggesting that temporal consistency cues are neglected or not fully utilized. In this paper, we set out tackle these issues, aim of achieving highly detailed and more temporall...

Video Instance Segmentation (VIS) is a task that simultaneously requires classification, segmentation, and instance association in video. Recent VIS approaches rely on sophisticated pipelines to achieve this goal, including RoI-related operations or 3D convolutions. In contrast, we present simple efficient single-stage framework based the segmentation method CondInst by adding an extra tracking...

Separating 3D point clouds into individual instances is an important task for vision. It challenging due to the unknown and varying number of in a scene. Existing deep learning based works focus on two-step pipeline: first learn feature embedding then cluster points. Such pipeline leads disconnected intermediate objectives. In this paper, we propose integrated reformulation instance segmentatio...

In this paper new techniques for optimizing instance retrieval in DL systems are described. The algorithms are evaluated with application examples from a natural language processing application.

We consider the problem of generating hard instances for the Satisfying Assignment Search Problem (in short, SAT). It is not known whether SAT is difficult on average, while it has been believed that the Factorization Problem (in short, FACT) is hard on average. Thus, one can expect to generate hard-on-average instances by using a reduction from FACT to SAT. Although the asymptotically best red...