Towards Unifying Reference Expression Generation and Comprehension

TL;DR

This paper introduces UniRef, a unified model for reference expression generation and comprehension that leverages a novel fusion layer and joint pre-training to improve performance on both tasks.

Contribution

The paper presents UniRef, a novel unified model with a specialized fusion layer and joint pre-training strategies for REG and REC tasks, addressing their interrelated challenges.

Findings

Outperforms previous state-of-the-art on REG and REC tasks

Effective fusion of image, region, and text improves task performance

Joint pre-training enhances the shared representation quality

Abstract

Reference Expression Generation (REG) and Comprehension (REC) are two highly correlated tasks. Modeling REG and REC simultaneously for utilizing the relation between them is a promising way to improve both. However, the problem of distinct inputs, as well as building connections between them in a single model, brings challenges to the design and training of the joint model. To address the problems, we propose a unified model for REG and REC, named UniRef. It unifies these two tasks with the carefully-designed Image-Region-Text Fusion layer (IRTF), which fuses the image, region and text via the image cross-attention and region cross-attention. Additionally, IRTF could generate pseudo input regions for the REC task to enable a uniform way for sharing the identical representation space across the REC and REG. We further propose Vision-conditioned Masked Language Modeling (VMLM) and Text-Conditioned Region Prediction (TRP) to pre-train UniRef model on multi-granular corpora. The VMLM and TRP are directly related to REG and REC, respectively, but could help each other. We conduct extensive experiments on three benchmark datasets, RefCOCO, RefCOCO+ and RefCOCOg. Experimental results show that our model outperforms previous state-of-the-art methods on both REG and REC.