From Show to Tell: A Survey on Deep Learning-Based Image Captioning.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Natural Language Processing

Background:

• Image captioning, crucial for generative intelligence, describes images with meaningful sentences.
• Early approaches used pipelines of visual encoders and language models, evolving with region exploitation and multi-modal connections.

Approach:

• This work provides a comprehensive overview of image captioning methods.
• It quantitatively compares state-of-the-art architectures and training strategies.
• The study discusses problem variants and open challenges.

Key Points:

• Significant advancements in visual encoding and language generation have been made.
• Fully-attentive and BERT-like early-fusion strategies have shown promise.
• Quantitative comparisons identify impactful technical innovations.

Conclusions:

• Despite progress, image captioning research is ongoing.
• This overview serves as a guide to existing literature and future research directions.
• The study highlights the synergy between Computer Vision and Natural Language Processing.