10 datasets found

Video instance segmentation on mobile devices is an impor- tant yet very challenging edge AI problem. It mainly suffers from (1) heavy computation and memory costs for frame-...

DAVIS2017 and DAVIS2017-Motion focus on the performance of instance segmentation on prominent moving objects.

YouTubeVIS-2019 and YouTubeVIS-2021 contain 2,883 high-resolution YouTube videos.

VideoCutLER is a simple unsupervised video instance segmentation method (UnVIS) that uses a cut-synthesis-and-learn pipeline to generate pseudo-masks for multiple objects in an...

Video instance segmentation requires consistently segmenting and tracking objects over time. Due to the quadratic dependency on input size, directly applying self-attention to...

The paper introduces an online method, named SAM-PD, that applies SAM to track and segment objects throughout the video.

The Youtube-VIS-2019 dataset, which is a benchmark for video instance segmentation.

A generalized framework for video instance segmentation.

A challenging video instance segmentation benchmark with object occlusions.

Video instance segmentation task requires classifying, segmenting instances in each frame, and tracking the same instance across frames.