In this research, we present image description generation using deep and transfer learning techniques for cross-domain images. The proposed system is able to conduct image captioning for out-of-scope images via fine-tuning and transfer learning. Tested with several cross-domain datasets, the work shows impressive capabilities for describing out-of-domain CCTV surveillance, artwork and healthcare images.
|Title of host publication||Data Science and Knowledge Engineering for Sensing Decision Support|
|Subtitle of host publication||Proceedings of the 13th International FLINS Conference (FLINS 2018)|
|Place of Publication||Northern Ireland|
|Publication status||Published - Oct 2018|