Hallucination in Large Vision-Language Models (LVLMs) remains a critical challenge, undermining their reliability in real-world applications. Existing studies have investigated the causes of hallucination at the modality level and proposed effective strategies. However, interaction patterns beyond the modality level remain insufficiently explored. In this paper, we conduct a token-level analysis and identify two key phenomena: (1) a small subset of textual tokens in LVLMs exert disproportionate influence in the visual-active layers, surpassing that of the visual modality and potentially misleading visual understanding; (2) while LVLMs can correctly identify key visual information, insufficient focus on these cues can sometimes lead to hallucinations. Based on such observation, we attribute hallucinations in LVLMs to two token-level causes: the disproportionate influence of certain textual tokens (phantom tokens) and the underutilization of critical visual cues (anchor tokens). To mitigate these issues, we introduce Token-Asymmetric Filtering (TAF)—a training-free, plug-and-play method that modulates intermediate attention maps in LVLMs. TAF isolates the influence of phantom tokens and emphasizes the influence of anchor tokens in the visual-active layers. Experimental results across multiple benchmarks demonstrate that TAF significantly mitigates hallucinations across a range of state-of-the-art LVLMs.

Semi-supervised learning (SSL) for medical image segmentation is a challenging yet highly practical task, which reduces reliance on large-scale labeled datasets by leveraging unlabeled samples. Among SSL techniques, the weak-to-strong consistency framework, popularized by FixMatch, has emerged as a state-of-the-art method in classification tasks. Notably, such a simple pipeline has also shown competitive performance in medical image segmentation. However, two key limitations still persist, impeding its efficient adaptation: (1) the neglect of contextual dependencies results in inconsistent predictions for similar semantic features, leading to incomplete object segmentation; (2) the lack of exploitation on semantic similarity between labeled and unlabeled data induces considerable class-distribution discrepancy. To address these limitations, we propose a novel SSL framework for medical image segmentation, named S2Match, powered by two appealing designs from a semantic similarity perspective: (1) rectifying pixel-wise prediction by reasoning about the intra-image pair-wise affinity map, thus integrating contextual dependencies explicitly into the final prediction; (2) bridging labeled and unlabeled data via a feature querying mechanism for compact class representation learning, which fully considers cross-image anatomical similarities. As the reliable semantic similarity extraction depends on robust features, we further introduce an effective Spatial-aware Fusion Module (SFM) to explore distinctive information from multiple scales. Experiments show that S2Match yields consistent improvements over the state-of-the-art methods across five public medical image segmentation benchmarks, exhibiting competitive performance on both 2D and 3D tasks.

The advancement of Spatial Transcriptomics (ST) has facilitated the spatially-aware profiling of gene expressions based on histopathology images. Although ST data offers valuable insights into the micro-environment of tumors, its acquisition cost remains expensive. Therefore, directly predicting the ST expressions from digital pathology images is desired. Current methods usually adopt existing regression backbones along with patch-sampling for this task, which ignores the inherent multi-scale information embedded in the pyramidal data structure of digital pathology images, and wastes the inter-spot visual information crucial for accurate gene expression prediction. To address these limitations, we propose M2OST, a many-to-one regression Transformer that can accommodate the hierarchical structure of the pathology images via a decoupled multi-scale feature extractor. Unlike traditional models that are trained with one-to-one image-label pairs, M2OST uses multiple images from different levels of the digital pathology image to jointly predict the gene expressions in their common corresponding spot. Built upon our many-to-one scheme, M2OST can be easily scaled to fit different numbers of inputs, and its network structure inherently incorporates nearby inter-spot features, enhancing regression performance. We have tested M2OST on three public ST datasets and the experimental results show that M2OST can achieve state-of-the-art performance with fewer parameters and floating-point operations (FLOPs).

Medical image segmentation is a crucial yet challenging task in image analysis across diverse anatomical structures. Current segmentation models heavily depend on large-scale datasets, which are laborious to collect and annotate. While generative models offer a promising alternative for data augmentation, most existing approaches are limited to single-modality outputs, either synthetic images or segmentation masks. Moreover, these methods often lack flexible conditioning mechanisms and struggle to capture the rich contextual dependencies inherent in anatomical structures. To address these challenges, in this paper, we propose TPCDM, a novel framework that co-synthesizes high-fidelity paired medical images and segmentation masks through a unified Triple-Prompt Conditional Diffusion Model. At the heart of TPCDM lies a newly defined joint image-label generation paradigm, termed Coordinated Distribution Learning, governed by three synergistic prompts: (1) a text prompt encoding global anatomical semantics; (2) a spatial prompt enforcing pixel-wise spatial coherence; (3) a task prompt dynamically adapting to diverse distributions. Furthermore, TPCDM disentangles instance-wise annotations into semantic masks and distance maps, enabling seamless extension to instance segmentation tasks. Extensive experiments on four benchmarks demonstrate that TPCDM achieves superior synthesis quality. Besides, incorporating the synthesized samples leads to state-of-the-art performance in both downstream semantic and instance segmentation tasks, while also delivering significant improvements under limited labeled data.

Multimodal sentiment analysis seeks to utilize various types of signals to identify underlying emotions and sentiments. A key challenge in this field lies in multimodal representation learning, which aims to develop effective methods for integrating multimodal features into cohesive representations. Recent advancements include two notable approaches: one focuses on decomposing multimodal features into modality-invariant and -specific components, while the other emphasizes the use of mutual information to enhance the fusion of modalities. Both strategies have demonstrated effectiveness and yielded remarkable results. In this paper, we propose a novel learning framework that combines the strengths of these two approaches, termed mutual information-based disentangled multimodal representation learning. Our approach involves estimating different types of information during feature extraction and fusion stages. Specifically, we quantitatively assess and adjust the proportions of modality-invariant, -specific, and -complementary information during feature extraction. Subsequently, during fusion, we evaluate the amount of information retained by each modality in the fused representation. We employ mutual information or conditional mutual information to estimate each type of information content. By reconciling the proportions of these different types of information, our approach achieves state-of-the-art performance on popular sentiment analysis benchmarks, including CMU-MOSI and CMU-MOSEI.

Multiple Instance Learning (MIL) has demonstrated promise in Whole Slide Image (WSI) classification. However, a major challenge persists due to the high computational cost associated with processing these gigapixel images. Existing methods generally adopt a two-stage approach, comprising a non-learnable feature embedding stage and a classifier training stage. Though it can greatly reduce memory consumption by using a fixed feature embedder pre-trained on other domains, such a scheme also results in a disparity between the two stages, leading to suboptimal classification accuracy. To address this issue, we propose that a bag-level classifier can be a good instance-level teacher. Based on this idea, we design Iteratively Coupled Multiple Instance Learning (ICMIL) to couple the embedder and the bag classifier at a low cost. ICMIL initially fixes the patch embedder to train the bag classifier, followed by fixing the bag classifier to fine-tune the patch embedder. The refined embedder can then generate better representations in return, leading to a more accurate classifier for the next iteration. To realize more flexible and more effective embedder fine-tuning, we also introduce a teacher-student framework to efficiently distill the category knowledge in the bag classifier to help the instance-level embedder fine-tuning. Intensive experiments were conducted on four distinct datasets to validate the effectiveness of ICMIL. The experimental results consistently demonstrated that our method significantly improves the performance of existing MIL backbones, achieving state-of-the-art results. The code and the organized datasets can be accessed by: https://github.com/Dootmaan/ICMIL/tree/confidence-based.

Conventional unsupervised domain adaptation (UDA) and domain generalization (DG) methods rely on the assumption that all source domains can be directly accessed and combined for model training. However, this centralized training strategy may violate privacy policies in many real-world applications. A paradigm for tackling this problem is to train multiple local models and aggregate a generalized central model without data sharing. Recent methods have made remarkable advancements in this paradigm by exploiting parameter alignment and aggregation. But when sources domain variety increases, directly aligning and aggregating local parameters becomes more challenging. Adapting a different approach in this work, we devised a data-free semantic collaborative distillation strategy to learn domain-invariant representation for both federated UDA and DG. Each local model transmits its predictions to the central server and derives its target distribution from the average of other local models' distributions to facilitate the mutual transfer of domain-specific knowledge. When unlabeled target data is available, we introduce a novel UDA strategy termed knowledge filter to adapt the central model to the target data. Extensive experiments on four UDA and DG datasets demonstrate that our method has a competitive performance compared with the state-of-the-art methods.

Referring image segmentation aims to segment an object out of an image via a specific language expression. The main concept is establishing global visual-linguistic relationships to locate the object and identify boundaries using details of the image. Recently, various Transformer-based techniques have been proposed to efficiently leverage long-range cross-modal dependencies, enhancing performance for referring segmentation. However, existing methods consider visual feature extraction and cross-modal fusion separately, resulting in insufficient visual-linguistic alignment in semantic space. In addition, they employ sequential structures and hence lack multi-scale information interaction. To address these limitations, we propose a Scale-Wise Language-Guided Vision Transformer (SLViT) with two appealing designs: (1) Language-Guided Multi-Scale Fusion Attention, a novel attention mechanism module for extracting rich local visual information and modeling global visual-linguistic relationships in an integrated manner. (2) An Uncertain Region Cross-Scale Enhancement module that can identify regions of high uncertainty using linguistic features and refine them via aggregated multi-scale features. We have evaluated our method on three benchmark datasets. The experimental results demonstrate that SLViT surpasses state-of-the-art methods with lower computational cost. The code is publicly available at: https://github.com/NaturalKnight/SLViT.

The task of multi-label image classification involves recognizing multiple objects within a single image. Considering both valuable semantic information contained in the labels and essential visual features presented in the image, tight visual-linguistic interactions play a vital role in improving classification performance. Moreover, given the potential variance in object size and appearance within a single image, attention to features of different scales can help to discover possible objects in the image. Recently, Transformer-based methods have achieved great success in multi-label image classification by leveraging the advantage of modeling long-range dependencies, but they have several limitations. Firstly, existing methods treat visual feature extraction and cross-modal fusion as separate steps, resulting in insufficient visual-linguistic alignment in the joint semantic space. Additionally, they only extract visual features and perform cross-modal fusion at a single scale, neglecting objects with different characteristics. To address these issues, we propose a Hierarchical Scale-Aware Vision-Language Transformer (HSVLT) with two appealing designs: (1)A hierarchical multi-scale architecture that involves a Cross-Scale Aggregation module, which leverages joint multi-modal features extracted from multiple scales to recognize objects of varying sizes and appearances in images. (2)Interactive Visual-Linguistic Attention, a novel attention mechanism module that tightly integrates cross-modal interaction, enabling the joint updating of visual, linguistic and multi-modal features. We have evaluated our method on three benchmark datasets. The experimental results demonstrate that HSVLT surpasses state-of-the-art methods with lower computational cost.

High resolution (HR) 3D medical image segmentation is vital for an accurate diagnosis. However, in the field of medical imaging, it is still a challenging task to achieve a high segmentation performance with cost-effective and feasible computation resources. Previous methods commonly use patch-sampling to reduce the input size, but this inevitably harms the global context and decreases the model's performance. In recent years, a few patch-free strategies have been presented to deal with this issue, but either they have limited performance due to their over-simplified model structures or they follow a complicated training process. In this study, to effectively address these issues, we present Adaptive Decomposition (A-Decomp) and Shared Weight Volumetric Transformer Blocks (SW-VTB). A-Decomp can adaptively decompose features and reduce their spatial size, which greatly lowers GPU memory consumption. SW-VTB is able to capture long-range dependencies at a low cost with its lightweight design and cross-scale weight-sharing mechanism. Our proposed cross-scale weight-sharing approach enhances the network's ability to capture scale-invariant core semantic information in addition to reducing parameter numbers. By combining these two designs together, we present a novel patch-free segmentation framework named VolumeFormer. Experimental results on two datasets show that VolumeFormer outperforms existing patch-based and patch-free methods with a comparatively fast inference speed and relatively compact design.

Though U-Net has achieved tremendous success in medical image segmentation tasks, it lacks the ability to explicitly model long-range dependencies. Therefore, Vision Transformers have emerged as alternative segmentation structures recently, for their innate ability of capturing long-range correlations through Self-Attention (SA). However, Transformers usually rely on large-scale pre-training and have high computational complexity. Furthermore, SA can only model self-affinities within a single sample, ignoring the potential correlations of the overall dataset. To address these problems, we propose a novel Transformer module named Mixed Transformer Module (MTM) for simultaneous inter- and intra- affinities learning. MTM first calculates self-affinities efficiently through our well-designed Local-Global Gaussian-Weighted Self-Attention (LGG-SA). Then, it mines inter-connections between data samples through External Attention (EA). By using MTM, we construct a U-shaped model named Mixed Transformer U-Net (MT-UNet) for accurate medical image segmentation. We test our method on two different public datasets, and the experimental results show that the proposed method achieves better performance over other state-of-the-art methods. The code is available at: https://github.com/Dootmaan/MT-UNet.

Multimodal sentiment analysis and depression estimation are two important research topics that aim to predict human mental states using multimodal data. Previous research has focused on developing effective fusion strategies for exchanging and integrating mind-related information from different modalities. Some MLP-based techniques have recently achieved considerable success in a variety of computer vision tasks. Inspired by this, we explore multimodal approaches with a feature-mixing perspective in this study. To this end, we introduce CubeMLP, a multimodal feature processing framework based entirely on MLP. CubeMLP consists of three independent MLP units, each of which has two affine transformations. CubeMLP accepts all relevant modality features as input and mixes them across three axes. After extracting the characteristics using CubeMLP, the mixed multimodal features are flattened for task predictions. Our experiments are conducted on sentiment analysis datasets: CMU-MOSI and CMU-MOSEI, and depression estimation dataset: AVEC2019. The results show that CubeMLP can achieve state-of-the-art performance with a much lower computing cost.