Semi Supervised Learning for Accurate Segmentation of Roughly Labeled Data


Rachel Rajan

Date of Award


Degree Name

M.S. in Electrical Engineering


Department of Electrical and Computer Engineering


Advisor: Vijayan Asari


Recent advancements in Neural Networks have obtained immense popularity in thefield of computer vision applications including image classification, semanticsegmentation, object detection and many more. Studies show that semantic segmentationhas always been a challenging task in computer vision. This requires a significantly numberof pixel-level annotated to assign a label to each image pixel. But, for supervised deeplearning techniques, the unavailability of labeled data has limited applications for accuratesemantic segmentation.Hence, an enhanced adversarial learning approach in semi-supervised segmentationis proposed for incremental training of the deep learning-based model to utilize unlabeleddata in achieving better learning performance. Studies reveal that unlabeled data combinedwith small amount of labeled data can improve the overall performance considerably. Sincemost of the existing methods use weakly labeled images, the proposed technique utilizesunlabeled instances to improve the segmentation model. A Generative and AdversarialNetwork (GAN) based semi-supervised framework is implemented here. This mainlyconsists of a generator and a discriminator, the generator provides extra training examplesto classifier, while the discriminator works on providing labels to the samples from the possible classes else assigns it as a pseudo label. The main motive of this implementationis to adding large pseudo labels turns the real samples to be closer in the feature spacehence improving the pixel level classification.Experiments on a publicly available datasets such as PASCAL VOC 2012 andPODOCYTE Benchmark dataset released by University at Buffalo, which demonstrate the effectiveness of the proposed method.


Computer Engineering, Semantic Segmentation, Semi-supervised Learning, Generative Adversarial Networks, Encoder-decoder, Computer Vision

Rights Statement

Copyright 2020, author