Recurrent Vision Transformer for Solving Visual Reasoning Problems

TL;DR

This paper introduces the Recurrent Vision Transformer (RViT), a model that combines recurrent connections and spatial attention to improve visual reasoning, achieving competitive results with fewer parameters and providing insights into attention and recurrence roles.

Contribution

The paper presents the RViT model that integrates recurrent connections with Transformers for visual reasoning, demonstrating its effectiveness and parameter efficiency compared to traditional CNNs.

Findings

RViT achieves competitive results on SVRT visual reasoning tasks.

Recurrent and attention mechanisms improve internal reasoning processes.

Model requires only 28k training samples for effective learning.

Abstract

Although convolutional neural networks (CNNs) showed remarkable results in many vision tasks, they are still strained by simple yet challenging visual reasoning problems. Inspired by the recent success of the Transformer network in computer vision, in this paper, we introduce the Recurrent Vision Transformer (RViT) model. Thanks to the impact of recurrent connections and spatial attention in reasoning tasks, this network achieves competitive results on the same-different visual reasoning problems from the SVRT dataset. The weight-sharing both in spatial and depth dimensions regularizes the model, allowing it to learn using far fewer free parameters, using only 28k training samples. A comprehensive ablation study confirms the importance of a hybrid CNN + Transformer architecture and the role of the feedback connections, which iteratively refine the internal representation until a stable prediction is obtained. In the end, this study can lay the basis for a deeper understanding of the role of attention and recurrent connections for solving visual abstract reasoning tasks.