An international team of researchers has created a dataset of videos with complex scenes to train video segmentation models to track a particular object throughout a video. The dataset contains about 2,150 videos featuring 5,200 objects from 36 categories and 431,725 object segmentation masks highlighting the relevant objects in the videos. The dataset improves upon past datasets by including complex scenes, such as crowded environments and objects that disappear and reappear.
Image credit: Flickr user pafotofan
