PointRend Segmentation for a Densely Occluded Moving Object in a Video
M Suresha, S. Kuppa, D. S. Raghukumar · 2021
Multimedia information retrieval segmenting of non-semantic and general objects is one of the important task. In particular video object segmentation has been added one more component called temporal, from this temporal component our objective is to find pixels which corresponding to the object in every consecutive frames. In these consecutive frame keenly observes its correlation and segment the instances without overlapping the neighbour objects values or boundary pixels, in this paper, we have elaborated on it. The key technique of our approach is to using PointRend classical subdivision techniques for point estimations and preserve the frame resolution. We will discuss a powerful method for segmenting the most densely occluded objects using PointRend segmentation in video, and we will show how we achieve competitive results on standard benchmark databases.