Deep-Learning-Based Approaches for Semantic Segmentation of Natural Scene Images: A Review

The task of semantic segmentation holds a fundamental position in the field computer vision. Assigning label to each pixel an image is challenging task. In recent times, significant advancements have been achieved through application Convolutional Neural Networks (CNN) techniques based on deep learning. This paper presents comprehensive and structured analysis approximately 150 methods CNN within last decade. Moreover, it examines 15 well-known datasets field. These consist 2D 3D video frames, including general, indoor, outdoor, street scenes. Furthermore, this mentions several techniques, such as SAM, UDA, common post-processing algorithms, CRF MRF. Additionally, analyzes performance evaluation reviewed state-of-the-art methods, pioneering backbone networks, popular datasets. compared according results Mean Intersection over Union (MIoU), most metric segmentation. Finally, discusses main challenges possible solutions underlines some future research directions We hope that our survey article will be useful provide foreknowledge readers who work