Sınıf-nitelik ilişkileri gerektirmeyen genelleştirilmiş sıfır-örnekli nesne tanıma

Over the last decade, great improvements have been achieved in image classification performances following the advances in supervised deep learning approaches. These supervised approaches, however, typically require substantial amounts of labeled training examples. Collecting and annotating such examples is a cumbersome and error-prone task, especially when a large number of classes needs to be spanned. One of the promising approaches towards overcoming this limitation of supervised recognition techniques is zero-shot learning. Inspired by the abilities of human vision, zero-shot learning aims to enable recognition of novel object categories purely based on category-wide information, which we refer to as auxiliary class information. A more modern variant, called generalized zero-shot learning, aims to build models that can accurately classify novel samples of not only zero-shot classes but also those with supervised training examples. Most of the recent generalized zero-shot learning approaches rely on attribute based auxiliary class information, where the attributes characterising each class of interest needs to be defined by an oracle. In practice, this dependency greatly reduces the practicality of zero-shot learning as it is often difficult to define such class-attribute relationships. To bypass this requirement, in this thesis, we propose a model that requires only class names of novel classes and implicitly learns pseudo-attributes in an end-to-end manner purely based on a set of candidate pseudo-attribute word embeddings. Such word embeddings are much easier to collect than class-attribute annotations, as one can easily select and utilize a set of relevant words from a pre-trained language model that provides vector-space word embeddings. Additionally, we propose a simple contrastive loss term for improving generalized zero-shot learning based on simple class-to-class name similarity scores. Our experimental results show that the proposed approach yields state-of-the-art class name based generalized zero-shot learning.