Few-Shot Semantic Segmentaion Using Meta-Learning<br/>

Please enable javascript in your browser.

Few-Shot Semantic Segmentaion Using Meta-Learning

394 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 54885 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Soleymani Baghshah, Mahdieh
Abstract:
Despite recent advancements in deep learning methods, these methods rely on a huge amount of training data to work. Recently the problem of solving classification and recently semantic segmentation problems with a few training data have gained attention to tackle this issue. In this research, we propose a meta-learning method by combining optimization-based and prototypical approaches in which a small portion of parameters are optimized with task-specific initialization. In addition to this and designing other parts of the method, we propose a new approach to use query data as an unlabeled sample to enhance task-specific learning. Alongside the mentioned method, we propose an approach to use probabilistic methods in meta-learning to take advantage of different types of statistics of support samples. Experiments on Pascal-5i show that using the proposed ideas leads to a 1.4% performance gain (MIoU) compared to prior works in 5-shot settings. Moreover on COCO-20i datasets the performance gain is 1.8% and 5.8% for 1-shot and 5-shot settings respectively.
Keywords:
Deep Learning ; Metalearning ; Semantic Segmentation ; Few-Shot Learning ; Image Segmentation