Continual semantic segmentation via image and latent space consistency

• Main Author: Wang, Zhichao
• Other Authors: Ponnuthurai Nagaratnam Suganthan
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/154665
• Institution: Nanyang Technological University

In this thesis, my continual-learning research process is introduced in detail, including a novel method and two regulators, which contribute to anti-forgetting Result in continual learning in the semantic segmentation area. Firstly a real-time semantic segmentation model called ERFnet is evaluated, then based on this network and Cityscapes dataset, a model-recall method is proposed which could significantly reduce the catastrophic forgetting which happens in the process of continual learning in the semantic segmentation area; inspired by mentors, 2 regulators are also conducted which were expected to further improve performance (one regulator is come up by mentors and another is by myself). A couple of experiments are designed to evaluate the performance of the new Idea and prediction images for each step is visible.