Understanding ME? Multimodal Evaluation for Fine-grained Visual Commonsense

TL;DR

This paper introduces a multimodal evaluation pipeline for visual commonsense understanding, revealing insights into model comprehension and demonstrating that training with generated data improves performance on standard benchmarks.

Contribution

It presents a novel automatic evaluation method for visual commonsense, and shows that training with generated data enhances model performance and understanding.

Findings

Visual information is underutilized compared to text.

Low-level semantic info aids high-level understanding.

Training with ME data improves VCR benchmark performance.

Abstract

Visual commonsense understanding requires Vision Language (VL) models to not only understand image and text but also cross-reference in-between to fully integrate and achieve comprehension of the visual scene described. Recently, various approaches have been developed and have achieved high performance on visual commonsense benchmarks. However, it is unclear whether the models really understand the visual scene and underlying commonsense knowledge due to limited evaluation data resources. To provide an in-depth analysis, we present a Multimodal Evaluation (ME) pipeline to automatically generate question-answer pairs to test models' understanding of the visual scene, text, and related knowledge. We then take a step further to show that training with the ME data boosts the model's performance in standard VCR evaluation. Lastly, our in-depth analysis and comparison reveal interesting findings: (1) semantically low-level information can assist the learning of high-level information but not the opposite; (2) visual information is generally under utilization compared with text.