A Deep Learning Model for More Descriptive Image Captioning
Tunahan Doner, Buket Kaya · 2024
Describing images with natural and fluent sentences is recognized as a challenging problem in computer vision and natural language processing. The limited number of studies in this area, especially in the Turkish language, and the fact that the existing datasets usually contain generalized descriptions made this research necessary. In this study, a rich dataset in Turkish language was created and this dataset was developed by optimizing both the natural structure of the language and the sentence lengths. An accuracy rate of 36% was achieved in the training process and 31% in the testing process. As a result, the generated dataset aims to contribute to Turkish image recognition models to produce more natural and detailed sentences.