Visual Dialogue without Vision or Dialogue

TL;DR

This paper introduces a simple CCA-based method for visual dialogue that ignores visual input and dialogue sequence, revealing issues in current complex models and achieving competitive results with minimal computation.

Contribution

The paper presents a minimalistic, gradient-free CCA approach that challenges existing complex models in visual dialogue by highlighting dataset biases and evaluation limitations.

Findings

CCA method achieves near state-of-the-art mean rank performance

Current models may rely on dataset biases rather than true understanding

Simpler models can perform competitively, questioning the necessity of complex architectures

Abstract

We characterise some of the quirks and shortcomings in the exploration of Visual Dialogue - a sequential question-answering task where the questions and corresponding answers are related through given visual stimuli. To do so, we develop an embarrassingly simple method based on Canonical Correlation Analysis (CCA) that, on the standard dataset, achieves near state-of-the-art performance on mean rank (MR). In direct contrast to current complex and over-parametrised architectures that are both compute and time intensive, our method ignores the visual stimuli, ignores the sequencing of dialogue, does not need gradients, uses off-the-shelf feature extractors, has at least an order of magnitude fewer parameters, and learns in practically no time. We argue that these results are indicative of issues in current approaches to Visual Dialogue and conduct analyses to highlight implicit dataset biases and effects of over-constrained evaluation metrics. Our code is publicly available.