Sardorbek Oromiddinov - Academia.edu (original) (raw)
Uploads
Papers by Sardorbek Oromiddinov
2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2020
Most existing ZSL models focus on searching the mapping between visual space and semantic space d... more Most existing ZSL models focus on searching the mapping between visual space and semantic space directly. However, few models study whether the human-designed semantic information is discriminative enough to recognize different categories. On the other hand, one-way mapping typically suffers from the project domain shift problem. Inspired by the encoder-decoder paradigm, we propose a novel solution to ZSL based on learning a Discriminative Dual Semantic Auto-encoder (DDSA). DDSA aims to build an aligned space to bridge the visual space and the semantic space by learning two bidirectional mappings, which provides us the required discriminative information about the visual and semantic features in the aligned space. The key to the proposed model is that we implicitly exact the principal information from visual and semantic space to construct aligned features, which is not only semantic-preserving but also discriminative. Extensive experiments on five benchmark data sets demonstrate the effectiveness of the proposed approach.
Theoretical & Applied Science, 2021
2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2020
Most existing ZSL models focus on searching the mapping between visual space and semantic space d... more Most existing ZSL models focus on searching the mapping between visual space and semantic space directly. However, few models study whether the human-designed semantic information is discriminative enough to recognize different categories. On the other hand, one-way mapping typically suffers from the project domain shift problem. Inspired by the encoder-decoder paradigm, we propose a novel solution to ZSL based on learning a Discriminative Dual Semantic Auto-encoder (DDSA). DDSA aims to build an aligned space to bridge the visual space and the semantic space by learning two bidirectional mappings, which provides us the required discriminative information about the visual and semantic features in the aligned space. The key to the proposed model is that we implicitly exact the principal information from visual and semantic space to construct aligned features, which is not only semantic-preserving but also discriminative. Extensive experiments on five benchmark data sets demonstrate the effectiveness of the proposed approach.
Theoretical & Applied Science, 2021