Multimodal large language models (MLLMs) have experienced significant advancements recently, but still struggle to recognize and interpret intricate details in high-resolution (HR) images effectively. While state-of-the-art (SOTA) MLLMs claim to process images at 4K resolution, existing MLLM benchmarks only support up to 2K, leaving the capabilities of SOTA models on true HR images largely untested. Furthermore, existing methods for enhancing HR image perception in MLLMs rely on computationally expensive visual instruction tuning. To address these limitations, we introduce HR-Bench, the first deliberately designed benchmark to rigorously evaluate MLLM performance on 4K & 8K images. Through extensive experiments, we demonstrate that while downsampling HR images leads to vision information loss, leveraging complementary modalities, e.g., text, can effectively compensate for this loss. Building upon this insight, we propose Divide, Conquer and Combine, a novel training-free framework for enhancing MLLM perception of HR images. Our method follows a three-staged approach: 1) Divide: recursively partitioning the HR image into patches and merging similar patches to minimize computational overhead, 2) Conquer: leveraging the MLLM to generate accurate textual descriptions for each image patch, and 3) Combine: utilizing the generated text descriptions to enhance the MLLM's understanding of the overall HR image. Extensive experiments show that: 1) the SOTA MLLM achieves 63% accuracy, which is markedly lower than the 87% accuracy achieved by humans on HR-Bench; 2) our method brings consistent and significant improvements (a relative increase of +6% on HR-Bench and +8% on general multimodal benchmarks).

In recent years, various distillation methods for semantic segmentation have been proposed. However, these methods typically train the student model to imitate the intermediate features or logits of the teacher model directly, thereby overlooking the high-discrepancy regions learned by both models, particularly the differences in instance edges. In this paper, we introduce a novel approach, called Difference-aware Distillation, to address this limitation. Our proposed method detects the discrepancies among the teacher model and the student model in the logit space through two masking mechanisms (i.e., masking by logit differences with respect to the ground truth labels and masking by differences in the predictive class probabilities), and guides the student model to restore the teacher's features with the focus on these highly-discrepant regions, resulting in improved segmentation performance. With the features jointly masked by these two mechanisms, the student model learns to preserve the teacher's features via a feature generation module, thus achieving better representation. Our experimental evaluation on three datasets, Cityscapes, Pascal2012, and ADE20 K, demonstrates our proposed approach outperforms several baselines considered. Further visualization analysis confirms that our method effectively directs the student model's attention to the discrepancies, such as the edges of small objects and the interiors of large objects.