PixLore: A Dataset-driven Approach to Rich Image Captioning

Image captioning, defined as the automatic generation of textual descriptions for images, stands in the middle of computer vision and natural language processing. Its relevance is not confined to academic exploration; it has tangible applications such as assisting visually impaired individuals in interpreting visual content, improving search engine capabilities, and enhancing content discoverability on digital platforms. While there have been notable strides in this field, the pursuit of models capable of producing human-like, contextually appropriate, and detailed captions continues. Early attempts in image captioning primarily relied on simple template-based methods [4], where captions were generated by filling in predefined templates with object, action, and attribute labels detected in the image. However, these methods lacked flexibility and often failed to capture the richness and diversity of natural language.