SiRi: A Simple Selective Retraining Mechanism for Transformer-based Visual Grounding

TL;DR

SiRi is a simple selective retraining method that improves transformer-based visual grounding by continually updating the encoder and periodically re-initializing other parameters, leading to significant performance gains.

Contribution

Introduces SiRi, a novel retraining mechanism that enhances vision-language transformer training by focusing on better encoder initialization and optimization.

Findings

Achieves 83.04% Top1 accuracy on RefCOCO+ testA, surpassing previous methods by over 10%.

Performs well even with limited training data.

Effective extension to other vision-language tasks.

Abstract

In this paper, we investigate how to achieve better visual grounding with modern vision-language transformers, and propose a simple yet powerful Selective Retraining (SiRi) mechanism for this challenging task. Particularly, SiRi conveys a significant principle to the research of visual grounding, i.e., a better initialized vision-language encoder would help the model converge to a better local minimum, advancing the performance accordingly. In specific, we continually update the parameters of the encoder as the training goes on, while periodically re-initialize rest of the parameters to compel the model to be better optimized based on an enhanced encoder. SiRi can significantly outperform previous approaches on three popular benchmarks. Specifically, our method achieves 83.04% Top1 accuracy on RefCOCO+ testA, outperforming the state-of-the-art approaches (training from scratch) by more than 10.21%. Additionally, we reveal that SiRi performs surprisingly superior even with limited training data. We also extend it to transformer-based visual grounding models and other vision-language tasks to verify the validity.