An Introduction to Deep Learning-Based Object Segmentation Methods

Object segmentation is a challenging task that requires both instance-level and pixel-level predictions, and it has widespread applications in autonomous driving, video analysis, scene understanding, and other fields. This paper first categorizes object segmentation algorithms, then conducts an in-depth analysis of some representative algorithms and their improved versions, discusses the advantages and disadvantages of these algorithms, and finally provides an outlook on the future development of object segmentation methods. This paper It is found that the top-down detection algorithm represented by MaskR-CNN has the disadvantages of large computation and poor real-time performance; Bottom-up methods have a large gap in SSAP dependent post-processing and so on, but some breakthroughs in some new algorithms, such as Deep snake, etc.; the direct segmentation methods represented by the SOLO series achieve the accuracy and speed of the balance, while methods such as YOLACT make up for the shortcomings in sample segmentation. This paper also identifies current research gaps and outlines future directions.