Large Multimodal Models as General In-Context Classifiers
This work addresses the problem of selecting appropriate multimodal models for classification, particularly for researchers and practitioners deciding between LMMs and VLMs, by demonstrating the competitive in-context learning capabilities of LMMs and proposing a method for open-world classification.
This paper investigates the use of Large Multimodal Models (LMMs) for classification, finding that while their zero-shot performance is lower than CLIP-like models, LMMs can match or surpass contrastive Vision-Language Models (VLMs) with a few in-context examples. For open-world classification, they propose CIRCLE, a training-free method that iteratively refines pseudo-labels for in-context examples, which establishes a robust baseline and surpasses VLM counterparts.
Which multimodal model should we use for classification? Previous studies suggest that the answer lies in CLIP-like contrastive Vision-Language Models (VLMs), due to their remarkable performance in zero-shot classification. In contrast, Large Multimodal Models (LMM) are more suitable for complex tasks. In this work, we argue that this answer overlooks an important capability of LMMs: in-context learning. We benchmark state-of-the-art LMMs on diverse datasets for closed-world classification and find that, although their zero-shot performance is lower than CLIP's, LMMs with a few in-context examples can match or even surpass contrastive VLMs with cache-based adapters, their "in-context" equivalent. We extend this analysis to the open-world setting, where the generative nature of LMMs makes them more suitable for the task. In this challenging scenario, LMMs struggle whenever provided with imperfect context information. To address this issue, we propose CIRCLE, a simple training-free method that assigns pseudo-labels to in-context examples, iteratively refining them with the available context itself. Through extensive experiments, we show that CIRCLE establishes a robust baseline for open-world classification, surpassing VLM counterparts and highlighting the potential of LMMs to serve as unified classifiers, and a flexible alternative to specialized models.