Visual Commonsense in Pretrained Unimodal and Multimodal Models
This addresses the problem of evaluating visual commonsense in AI models for researchers, but it is incremental as it focuses on benchmarking rather than proposing new methods.
The paper investigates how well unimodal and multimodal models capture visually salient attributes by creating the ViComTe dataset covering 5 property types for over 5000 subjects, finding that multimodal models better reconstruct attribute distributions but are still affected by reporting bias, with model size not improving performance.
Our commonsense knowledge about objects includes their typical visual attributes; we know that bananas are typically yellow or green, and not purple. Text and image corpora, being subject to reporting bias, represent this world-knowledge to varying degrees of faithfulness. In this paper, we investigate to what degree unimodal (language-only) and multimodal (image and language) models capture a broad range of visually salient attributes. To that end, we create the Visual Commonsense Tests (ViComTe) dataset covering 5 property types (color, shape, material, size, and visual co-occurrence) for over 5000 subjects. We validate this dataset by showing that our grounded color data correlates much better than ungrounded text-only data with crowdsourced color judgments provided by Paik et al. (2021). We then use our dataset to evaluate pretrained unimodal models and multimodal models. Our results indicate that multimodal models better reconstruct attribute distributions, but are still subject to reporting bias. Moreover, increasing model size does not enhance performance, suggesting that the key to visual commonsense lies in the data.