Rewarding fine-grained image captioning with keyword group contrastive

Key Points

• Improved fine-grained image captioning accuracy was observed using a keyword contrastive method.
• The method relies on semantic similarity to enhance image captioning, reporting a significant increase in effectiveness.
• This analysis employed a machine learning framework to benchmark the performance of image captioning models.
• Highlighting the approach's potential, the findings suggest advancements in fine-grained understanding could enhance user experience.