Print Email Facebook Twitter Learn representations in the presence of segmentation label noises Title Learn representations in the presence of segmentation label noises Author Ju, Jihong (TU Delft Electrical Engineering, Mathematics and Computer Science) Contributor van Gemert, J.C. (mentor) Loog, M. (mentor) Szlávik, Zoltan (mentor) Hanjalic, A. (graduation committee) Degree granting institution Delft University of Technology Date 2017-08-31 Abstract Training data for segmentation tasks are often available only on a small scale. Transferring learned representations from pre-trained classification models is therefore widely adopted by convolutional neural networks for semantic segmentation. In domains where the representations from the classification models are not directly applicable, we propose to train representations with segmentation datasets that potentially contains label errors. Our experiments demonstrate that label errors, such as mislabeled segments and missing segmentations, have negative influences to the learned representations. To alleviate the negative effects of object mislabelling, we propose to discard the object labels and instead train foreground/background segmentation. The learned representations with binary segmentation achieve a fine-tuning performance comparable to the representations learned with ``gold'' standard segmentations. In the existence of missing segmentations, a sigmoid loss for the background class is proposed to achieve high recall while keeping the precision better than simply weighting the classes. The proposed class dependent, sigmoid loss obtains better segmentation performance as well as better representations than the weighting the classes in the presence of missing segmentations. To summerize, we propose to learn representations with foreground/background segmentation and with a sigmoid loss for the background class when there exist missing segmentations for objects. Subject Transfer learningImage segmentationPU learning To reference this document use: http://resolver.tudelft.nl/uuid:202633b6-4fb1-463a-a29e-b2f3e2402c00 Part of collection Student theses Document type master thesis Rights © 2017 Jihong Ju Files PDF thesis_jihong_ju_4518454.pdf 2.93 MB Close viewer /islandora/object/uuid:202633b6-4fb1-463a-a29e-b2f3e2402c00/datastream/OBJ/view