Deeper Thought, Weaker Aim: Understanding and Mitigating Perceptual Impairment during Reasoning in Multimodal Large Language Models

Multimodal large language models (MLLMs) often suffer from perceptual impairments under extended reasoning modes, particularly in visual question answering (VQA) tasks. We identify attention dispersion as the underlying cause: during multi-step reasoning, the model's visual attention becomes scattered and drifts away from question-relevant regions, effectively "losing focus" on the visual input. To better understand this phenomenon, we analyze the attention maps of MLLMs and observe that reasoning prompts significantly reduce attention to regions critical for answering the question. We further find a strong correlation between the model's overall attention on image tokens and the spatial dispersiveness of its attention within the image. Leveraging this insight, we propose a training-free Visual Region-Guided Attention (VRGA) framework that selects visual heads based on an entropy-focus criterion and reweights their attention, effectively guiding the model to focus on question-relevant regions during reasoning. Extensive experiments on vision-language benchmarks demonstrate that our method effectively alleviates perceptual degradation, leading to improvements in visual grounding and reasoning accuracy while providing interpretable insights into how MLLMs process visual information.

Iterative-in-Iterative Super-Resolution Biomedical Imaging Using One Real Image

Deep learning-based super-resolution models have the potential to revolutionize biomedical imaging and diagnoses by effectively tackling various challenges associated with early detection, personalized medicine, and clinical automation. However, the requirement of an extensive collection of high-resolution images presents limitations for widespread adoption in clinical practice. In our experiment, we proposed an approach to effectively train the deep learning-based super-resolution models using only one real image by leveraging self-generated high-resolution images. We employed a mixed metric of image screening to automatically select images with a distribution similar to ground truth, creating an incrementally curated training data set that encourages the model to generate improved images over time. After five training iterations, the proposed deep learning-based super-resolution model experienced a 7.5\% and 5.49\% improvement in structural similarity and peak-signal-to-noise ratio, respectively. Significantly, the model consistently produces visually enhanced results for training, improving its performance while preserving the characteristics of original biomedical images. These findings indicate a potential way to train a deep neural network in a self-revolution manner independent of real-world human data.

Self-similarity-based super-resolution of photoacoustic angiography from hand-drawn doodles

Deep-learning-based super-resolution photoacoustic angiography (PAA) is a powerful tool that restores blood vessel images from under-sampled images to facilitate disease diagnosis. Nonetheless, due to the scarcity of training samples, PAA super-resolution models often exhibit inadequate generalization capabilities, particularly in the context of continuous monitoring tasks. To address this challenge, we propose a novel approach that employs a super-resolution PAA method trained with forged PAA images. We start by generating realistic PAA images of human lips from hand-drawn curves using a diffusion-based image generation model. Subsequently, we train a self-similarity-based super-resolution model with these forged PAA images. Experimental results show that our method outperforms the super-resolution model trained with authentic PAA images in both original-domain and cross-domain tests. Specially, our approach boosts the quality of super-resolution reconstruction using the images forged by the deep learning model, indicating that the collaboration between deep learning models can facilitate generalization, despite limited initial dataset. This approach shows promising potential for exploring zero-shot learning neural networks for vision tasks.