BROWSE
Detail
Conference Paper
Word-piece based natural language recognition for multimodal object retrieval
Cited 0 time in 
Share 
Share
- Authors
- Joonyoung Jung, Dong-oh Kang
- Issue Date
- 2023-10
- Citation
- International Conference on Control, Automation and Systems (ICCAS 2023), pp.905-908
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- This paper proposes a word-piece based natural language recognition for multimodal object retrieval. Efforts are being made to enable flexible interaction between humans and robots by incorporating image and natural language-related multimodal AI technologies. This includes research on retrieving appropriate images even when natural language does not explicitly include words representing images. However, the lack of clear words in natural language corresponding to images poses challenges in multimodal object retrieval. In this paper, a proposed approach aims to enhance the performance of multimodal object retrieval through word-piece embedding-based natural language recognition (NLR). An experiment was conducted on multimodal object retrieval using word-piece embedding-based NLR. It demonstrated superior performance compared to word embedding. For example, with a data size of 1070, the top-2 retrieval accuracy achieved through word-piece embedding-based NLR was 0.75, while the top-2 retrieval accuracy achieved through word embedding-based NLR was 0.69.
- KSP Keywords
- Data size, Flexible interaction, Object retrieval, Word Embedding, language recognition, natural language, retrieval accuracy, superior performance