A Review on Deep Learning based Image Caption Generation Methods

This review investigates recent advancements in Deep Learning based image caption generation methodologies. The progression from early encoder-decoder models to attention-based systems and transformer frameworks is examined, alongside notable datasets and evaluation metrics. Strengths and shortcomings of leading methods are analyzed to provide researchers with a consolidated understanding of current trends. The proposed conceptual model integrates Convolutional Neural Networks for visual feature extraction with transformer mechanisms for sequence generation, aiming to produce accurate, fluent, and context-aware captions. Comparative discussions of state-of-the-art techniques are supported with evaluations based on benchmark datasets like MS- COCO and Flickr8k, emphasizing attention strategies including soft, hard, and adaptive variations.