This paper presents a method for generating and comprehending natural language referring expressions for objects in images. The authors focus on incorporating better measures of visual context into referring expression models, finding that visual comparisons to other objects within an image significantly improve performance. They also develop methods to tie the language generation process together, generating expressions for all objects of a particular category jointly. Evaluation on three recent datasets—RefCOCO, RefCOCO+, and RefCOCOg—shows the advantages of their methods for both referring expression generation and comprehension. The paper discusses the importance of visual comparisons in generating unambiguous referring expressions. It introduces a model that explicitly encodes the visual differences between objects of the same category, which helps in distinguishing the target object from others. Additionally, the authors propose a method to jointly generate referring expressions for all objects of the same category, ensuring that expressions are both distinct and complementary. The authors also explore the use of LSTM networks for language generation, incorporating visual and linguistic comparisons to improve performance. They compare their model with existing approaches and show that their model outperforms previous state-of-the-art methods in both generation and comprehension tasks. The experiments demonstrate that their model achieves higher accuracy in referring expression comprehension and generates more accurate and distinct expressions for referring to objects in images. The results indicate that incorporating visual comparisons and jointly generating expressions for multiple objects improves the performance of referring expression models.This paper presents a method for generating and comprehending natural language referring expressions for objects in images. The authors focus on incorporating better measures of visual context into referring expression models, finding that visual comparisons to other objects within an image significantly improve performance. They also develop methods to tie the language generation process together, generating expressions for all objects of a particular category jointly. Evaluation on three recent datasets—RefCOCO, RefCOCO+, and RefCOCOg—shows the advantages of their methods for both referring expression generation and comprehension. The paper discusses the importance of visual comparisons in generating unambiguous referring expressions. It introduces a model that explicitly encodes the visual differences between objects of the same category, which helps in distinguishing the target object from others. Additionally, the authors propose a method to jointly generate referring expressions for all objects of the same category, ensuring that expressions are both distinct and complementary. The authors also explore the use of LSTM networks for language generation, incorporating visual and linguistic comparisons to improve performance. They compare their model with existing approaches and show that their model outperforms previous state-of-the-art methods in both generation and comprehension tasks. The experiments demonstrate that their model achieves higher accuracy in referring expression comprehension and generates more accurate and distinct expressions for referring to objects in images. The results indicate that incorporating visual comparisons and jointly generating expressions for multiple objects improves the performance of referring expression models.