Advertisement
Published in:
2020 | OriginalPaper | Chapter
Enhanced Hybrid Segmentation with Non Local Block and Deep Residual Networks
Authors : Yuvaram Singh, Guda Ramachandra Kaladhara Sarma, Kameshwar Rao
Published in: Modeling, Machine Learning and Astronomy
Publisher: Springer Singapore
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Abstract
The use of convolutional neural networks (CNNs) has increased in the edge devices due to its successful performance. Various such applications includes semantic segmentation which is one of the most challenging tasks due to the involvement of tremendous model size and parameters. In this paper, an enhanced hybrid segmentation with non-local block and deep residual networks is introduced for pixel level semantic segmentation. A light weight model is developed to facilitate deployment on edge devices. Skip connection is applied to fire layer in encoder and decoder block of segmentation model and non-local block is inserted in between encoder and decoder. Due to such amendments, the proposed network has optimized the number of parameters to only 2 Million whereas SegNet-Basic architecture required 5 Million. The performance is validated on the Camvid dataset and 87.5% accuracy is achieved.