Generation and Comprehension of Unambiguous Object Descriptions

TLDR

The task of generating and interpreting unambiguous referring expressions is motivated by recent deep‑learning image captioning successes but offers an objectively evaluable alternative to the inherently ambiguous captioning problem. The authors aim to develop a method that both generates and interprets unambiguous referring expressions and to provide a large‑scale dataset for this task. The approach uses a deep‑learning model inspired by image captioning to generate and interpret referring expressions, supported by a newly released large‑scale dataset and evaluation toolbox. The method outperforms prior approaches that ignore scene ambiguity, and the authors have released the dataset and a visualization/evaluation toolbox.

Abstract

We propose a method that can generate an unambiguous description (known as a referring expression) of a specific object or region in an image, and which can also comprehend or interpret such an expression to infer which object is being described. We show that our method outperforms previous methods that generate descriptions of objects without taking into account other potentially ambiguous objects in the scene. Our model is inspired by recent successes of deep learning methods for image captioning, but while image captioning is difficult to evaluate, our task allows for easy objective evaluation. We also present a new large-scale dataset for referring expressions, based on MS-COCO. We have released the dataset and a toolbox for visualization and evaluation, see https://github.com/mjhucla/Google_Refexp_toolbox