Multimodal large language models (MLLMs) have shown impressive capabilities in document understanding, a rapidly growing research area with significant industrial demand. As a multimodal task, document understanding requires models to possess both perceptual and cognitive abilities. However, due to different types of annotation noise in training, current MLLMs often face conflicts between perception and cognition. Taking a document VQA task (cognition) as an example, an MLLM might generate answers that do not match the corresponding visual content identified by its OCR (perception). This conflict suggests that the MLLM might struggle to establish an intrinsic connection between the information it “sees” and what it “understands”. Such conflicts challenge the intuitive notion that cognition is consistent with perception, hindering the performance and explainability of MLLMs. In this paper, we define the conflicts between cognition and perception as Cognition and Perception (C&P) knowledge conflicts, a form of multimodal knowledge conflicts, and systematically assess them with a focus on document understanding. Our analysis reveals that even GPT-4o, a leading MLLM, achieves only 75.26% C&P consistency. To mitigate the C&P knowledge conflicts, we propose a novel method called Multimodal Knowledge Consistency Fine-tuning. Our method reduces C&P knowledge conflicts across all tested MLLMs and enhances their performance in both cognitive and perceptual tasks.

Conventional visual relationship detection models only use the numeric ids of relation labels for training, but ignore the semantic correlation between the labels, which leads to severe training biases and harms the generalization ability of representations. In this paper, we introduce compact language information of relation labels for regularizing the representation learning of visual relations. Specifically, we propose a simple yet effective visual Relationship prediction framework that transfers natural language knowledge learned from Contrastive Language-Image Pre-training (CLIP) models to enhance the relationship prediction, termed RelCLIP. Benefiting from the powerful visual-semantic alignment ability of CLIP at image level, we introduce a novel Relational Contrastive Learning (RCL) approach which explores relation-level visual-semantic alignment via learning to match cross-modal relational embeddings. By collaboratively learning the semantic coherence and discrepancy from relation triplets, the model can generate more discriminative and robust representations. Experimental results on the Visual Genome dataset show that RelCLIP achieves significant improvements over strong baselines under full (provide accurate labels) and distant supervision (provide noise labels), demonstrating its powerful generalization ability in learning relationship representations. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/RelCLIP.