Project Overview
This project addresses the challenge of automatic image captioning in Turkish, a language with different linguistic characteristics compared to English. The work extends the MS COCO dataset with Turkish captions and evaluates state-of-the-art models.
Key Features
- Turkish MS COCO Dataset: Extended dataset with Turkish language captions
- Deep Learning Models: Implementation of transformer-based architectures
- Multilingual NLP: Natural language processing for Turkish language
- Performance Evaluation: Comprehensive benchmarking of captioning models
Technologies Used
- Python
- TensorFlow/PyTorch
- Transformers
- Computer Vision libraries
- Natural Language Processing tools
Research Impact
This work achieved state-of-the-art performance for Turkish image captioning with a Bleu-1 score of 0.72 using Meshed Memory Transformers. The dataset serves as a valuable resource for Turkish NLP research and machine translation studies.
Publication
This work has been published in the 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME).