The increasing demands in analyzing complex associated scenes pose necessities to researching multi-image understanding abilities. Compared with understanding individual images, both the alignments and differences between images are essential aspects of understanding the intricate relationships for multi-image inference tasks. However, existing benchmarks face difficulties in addressing both of these aspects simultaneously, resulting in obstacles to modeling relationships under various granularities and domains of images. In this paper, we introduce M4Bench to enhance the capability of aligning and distinguishing multi-images with multi-domain multi-granularity comparison. We carefully design five comparison tasks related to coarse and fine-grained granularities in single and multiple domains of images and evaluate them on 13 state-of-the-art multi-modal large language models with various sizes. Besides, we analyze the evaluation results and provide several observations and viewpoints for the multi-image understanding research. The data and evaluation code are available at https://github.com/eaglelab-zju/M4Bench.
Multimodal Large Language Models (MLLMs) have demonstrated strong performance in visual understanding tasks, yet they often suffer from object hallucinations--generating descriptions of objects that are inconsistent with or entirely absent from the input. This issue is closely related to dataset biases, where frequent co-occurrences of objects lead to entangled semantic representations across modalities. As a result, models may erroneously activate object representations that are commonly associated with the input but not actually present. To address this, we propose a causality-driven disentanglement framework that mitigates hallucinations through causal intervention. Our approach includes a Causal-Driven Projector in the visual pathway and a Causal Intervention Module integrated into the final transformer layer of the language model. These components work together to reduce spurious correlations caused by biased training data. Experimental results show that our method significantly reduces hallucinations while maintaining strong performance on multiple multimodal benchmarks. Visualization analyses further confirm improved separability of object representations. The code is available at: https://github.com/IgniSavium/Causal-LLaVA