Learning Visual Reasoning Without Strong Priors

TL;DR

This paper demonstrates that a general-purpose Conditional Batch Normalization approach can achieve state-of-the-art visual reasoning performance on CLEVR without specialized architectures, learning a question-dependent multi-step reasoning process.

Contribution

It introduces a simple, general architecture using Conditional Batch Normalization that effectively learns visual reasoning without strong priors or hand-crafted modules.

Findings

Achieves 2.4% error on CLEVR benchmark, outperforming previous methods.

Learns a question-dependent, multi-step reasoning process.

Does not rely on specialized reasoning architectures.

Abstract

Achieving artificial visual reasoning - the ability to answer image-related questions which require a multi-step, high-level process - is an important step towards artificial general intelligence. This multi-modal task requires learning a question-dependent, structured reasoning process over images from language. Standard deep learning approaches tend to exploit biases in the data rather than learn this underlying structure, while leading methods learn to visually reason successfully but are hand-crafted for reasoning. We show that a general-purpose, Conditional Batch Normalization approach achieves state-of-the-art results on the CLEVR Visual Reasoning benchmark with a 2.4% error rate. We outperform the next best end-to-end method (4.5%) and even methods that use extra supervision (3.1%). We probe our model to shed light on how it reasons, showing it has learned a question-dependent, multi-step process. Previous work has operated under the assumption that visual reasoning calls for a specialized architecture, but we show that a general architecture with proper conditioning can learn to visually reason effectively.