This study proposes a novel few-shot crack image classification model based on CLIP and Bayesian optimization. By combining multimodal information and Bayesian approach, the model achieves efficient classification of crack images in a small number of training samples. The CLIP model employs its robust feature extraction capabilities to facilitate precise classification with a limited number of samples. In contrast, Bayesian optimisation enhances the robustness and generalization of the model, while reducing the reliance on extensive labelled data. The results demonstrate that the model exhibits robust performance across a diverse range of dataset scales, particularly in the context of small sample sets. The study validates the potential of the method in civil engineering crack classification.
Despite significant progress in Vision-Language Pre-training (VLP), current approaches predominantly emphasize feature extraction and cross-modal comprehension, with limited attention to generating or transforming visual content. This gap hinders the model's ability to synthesize coherent and novel visual representations from textual prompts, thereby reducing the effectiveness of multi-modal learning. In this work, we propose MedUnifier, a unified VLP framework tailored for medical data. MedUnifier seamlessly integrates text-grounded image generation capabilities with multi-modal learning strategies, including image-text contrastive alignment, image-text matching and image-grounded text generation. Unlike traditional methods that reply on continuous visual representations, our approach employs visual vector quantization, which not only facilitates a more cohesive learning strategy for cross-modal understanding but also enhances multi-modal generation quality by effectively leveraging discrete representations. Our framework's effectiveness is evidenced by the experiments on established benchmarks, including uni-modal tasks (supervised fine-tuning), cross-modal tasks (image-text retrieval and zero-shot image classification), and multi-modal tasks (medical report generation, image synthesis), where it achieves state-of-the-art performance across various tasks. MedUnifier also offers a highly adaptable tool for a wide range of language and vision tasks in healthcare, marking advancement toward the development of a generalizable AI model for medical applications.
Thyroid scintigraphy is a key imaging modality for diagnosing thyroid disorders. Deep learning models for thyroid scintigraphy classification often face challenges due to limited and imbalanced datasets, leading to suboptimal generalization. In this study, we investigate the effectiveness of different data augmentation techniques including Stable Diffusion (SD), Flow Matching (FM), and Conventional Augmentation (CA) to enhance the performance of a ResNet18 classifier for thyroid condition classification. Our results showed that FM-based augmentation consistently outperforms SD-based approaches, particularly when combined with original (O) data and CA (O+FM+CA), achieving both high accuracy and fair classification across Diffuse Goiter (DG), Nodular Goiter (NG), Normal (NL), and Thyroiditis (TI) cases. The Wilcoxon statistical analysis further validated the superiority of O+FM and its variants (O+FM+CA) over SD-based augmentations in most scenarios. These findings highlight the potential of FM-based augmentation as a superior approach for generating high-quality synthetic thyroid scintigraphy images and improving model generalization in medical image classification.
Although multi-instance learning (MIL) has succeeded in pathological image classification, it faces the challenge of high inference costs due to processing numerous patches from gigapixel whole slide images (WSIs). To address this, we propose HDMIL, a hierarchical distillation multi-instance learning framework that achieves fast and accurate classification by eliminating irrelevant patches. HDMIL consists of two key components: the dynamic multi-instance network (DMIN) and the lightweight instance pre-screening network (LIPN). DMIN operates on high-resolution WSIs, while LIPN operates on the corresponding low-resolution counterparts. During training, DMIN are trained for WSI classification while generating attention-score-based masks that indicate irrelevant patches. These masks then guide the training of LIPN to predict the relevance of each low-resolution patch. During testing, LIPN first determines the useful regions within low-resolution WSIs, which indirectly enables us to eliminate irrelevant regions in high-resolution WSIs, thereby reducing inference time without causing performance degradation. In addition, we further design the first Chebyshev-polynomials-based Kolmogorov-Arnold classifier in computational pathology, which enhances the performance of HDMIL through learnable activation layers. Extensive experiments on three public datasets demonstrate that HDMIL outperforms previous state-of-the-art methods, e.g., achieving improvements of 3.13% in AUC while reducing inference time by 28.6% on the Camelyon16 dataset.
Top-down attention plays a crucial role in the human vision system, wherein the brain initially obtains a rough overview of a scene to discover salient cues (i.e., overview first), followed by a more careful finer-grained examination (i.e., look closely next). However, modern ConvNets remain confined to a pyramid structure that successively downsamples the feature map for receptive field expansion, neglecting this crucial biomimetic principle. We present OverLoCK, the first pure ConvNet backbone architecture that explicitly incorporates a top-down attention mechanism. Unlike pyramid backbone networks, our design features a branched architecture with three synergistic sub-networks: 1) a Base-Net that encodes low/mid-level features; 2) a lightweight Overview-Net that generates dynamic top-down attention through coarse global context modeling (i.e., overview first); and 3) a robust Focus-Net that performs finer-grained perception guided by top-down attention (i.e., look closely next). To fully unleash the power of top-down attention, we further propose a novel context-mixing dynamic convolution (ContMix) that effectively models long-range dependencies while preserving inherent local inductive biases even when the input resolution increases, addressing critical limitations in existing convolutions. Our OverLoCK exhibits a notable performance improvement over existing methods. For instance, OverLoCK-T achieves a Top-1 accuracy of 84.2%, significantly surpassing ConvNeXt-B while using only around one-third of the FLOPs/parameters. On object detection, our OverLoCK-S clearly surpasses MogaNet-B by 1% in AP^b. On semantic segmentation, our OverLoCK-T remarkably improves UniRepLKNet-T by 1.7% in mIoU. Code is publicly available at https://rb.gy/wit4jh.
Vision-language models (VLMs) have made significant progress in image classification by training with large-scale paired image-text data. Their performances largely depend on the prompt quality. While recent methods show that visual descriptions generated by large language models (LLMs) enhance the generalization of VLMs, class-specific prompts may be inaccurate or lack discrimination due to the hallucination in LLMs. In this paper, we aim to find visually discriminative prompts for fine-grained categories with minimal supervision and no human-in-the-loop. An evolution-based algorithm is proposed to progressively optimize language prompts from task-specific templates to class-specific descriptions. Unlike optimizing templates, the search space shows an explosion in class-specific candidate prompts. This increases prompt generation costs, iterative times, and the overfitting problem. To this end, we first introduce several simple yet effective edit-based and evolution-based operations to generate diverse candidate prompts by one-time query of LLMs. Then, two sampling strategies are proposed to find a better initial search point and reduce traversed categories, saving iteration costs. Moreover, we apply a novel fitness score with entropy constraints to mitigate overfitting. In a challenging one-shot image classification setting, our method outperforms existing textual prompt-based methods and improves LLM-generated description methods across 13 datasets. Meanwhile, we demonstrate that our optimal prompts improve adapter-based methods and transfer effectively across different backbones.
Prompt tuning has become a popular strategy for adapting Vision-Language Models (VLMs) to zero/few-shot visual recognition tasks. Some prompting techniques introduce prior knowledge due to its richness, but when learnable tokens are randomly initialized and disconnected from prior knowledge, they tend to overfit on seen classes and struggle with domain shifts for unseen ones. To address this issue, we propose the InPK model, which infuses class-specific prior knowledge into the learnable tokens during initialization, thus enabling the model to explicitly focus on class-relevant information. Furthermore, to mitigate the weakening of class information by multi-layer encoders, we continuously reinforce the interaction between learnable tokens and prior knowledge across multiple feature levels. This progressive interaction allows the learnable tokens to better capture the fine-grained differences and universal visual concepts within prior knowledge, enabling the model to extract more discriminative and generalized text features. Even for unseen classes, the learned interaction allows the model to capture their common representations and infer their appropriate positions within the existing semantic structure. Moreover, we introduce a learnable text-to-vision projection layer to accommodate the text adjustments, ensuring better alignment of visual-text semantics. Extensive experiments on 11 recognition datasets show that InPK significantly outperforms state-of-the-art methods in multiple zero/few-shot image classification tasks.
Masked autoencoders (MAEs) represent a prominent self-supervised learning paradigm in computer vision. Despite their empirical success, the underlying mechanisms of MAEs remain insufficiently understood. Recent studies have attempted to elucidate the functioning of MAEs through contrastive learning and feature representation analysis, yet these approaches often provide only implicit insights. In this paper, we propose a new perspective for understanding MAEs by leveraging the information bottleneck principle in information theory. Our theoretical analyses reveal that optimizing the latent features to balance relevant and irrelevant information is key to improving MAE performance. Building upon our proofs, we introduce MI-MAE, a novel method that optimizes MAEs through mutual information maximization and minimization. By enhancing latent features to retain maximal relevant information between them and the output, and minimizing irrelevant information between them and the input, our approach achieves better performance. Extensive experiments on standard benchmarks show that MI-MAE significantly outperforms MAE models in tasks such as image classification, object detection, and semantic segmentation. Our findings validate the theoretical framework and highlight the practical advantages of applying the information bottleneck principle to MAEs, offering deeper insights for developing more powerful self-supervised learning models.
Although efficient extraction of discriminative spatial-spectral features is critical for hyperspectral images classification (HSIC), it is difficult to achieve these features due to factors such as the spatial-spectral heterogeneity and noise effect. This paper presents a Spatial-Spectral Diffusion Contrastive Representation Network (DiffCRN), based on denoising diffusion probabilistic model (DDPM) combined with contrastive learning (CL) for HSIC, with the following characteristics. First,to improve spatial-spectral feature representation, instead of adopting the UNets-like structure which is widely used for DDPM, we design a novel staged architecture with spatial self-attention denoising module (SSAD) and spectral group self-attention denoising module (SGSAD) in DiffCRN with improved efficiency for spectral-spatial feature learning. Second, to improve unsupervised feature learning efficiency, we design new DDPM model with logarithmic absolute error (LAE) loss and CL that improve the loss function effectiveness and increase the instance-level and inter-class discriminability. Third, to improve feature selection, we design a learnable approach based on pixel-level spectral angle mapping (SAM) for the selection of time steps in the proposed DDPM model in an adaptive and automatic manner. Last, to improve feature integration and classification, we design an Adaptive weighted addition modul (AWAM) and Cross time step Spectral-Spatial Fusion Module (CTSSFM) to fuse time-step-wise features and perform classification. Experiments conducted on widely used four HSI datasets demonstrate the improved performance of the proposed DiffCRN over the classical backbone models and state-of-the-art GAN, transformer models and other pretrained methods. The source code and pre-trained model will be made available publicly.
Domain Generalization (DG) research has gained considerable traction as of late, since the ability to generalize to unseen data distributions is a requirement that eludes even state-of-the-art training algorithms. In this paper we observe that the initial iterations of model training play a key role in domain generalization effectiveness, since the loss landscape may be significantly different across the training and test distributions, contrary to the case of i.i.d. data. Conflicts between gradients of the loss components of each domain lead the optimization procedure to undesirable local minima that do not capture the domain-invariant features of the target classes. We propose alleviating domain conflicts in model optimization, by iteratively annealing the parameters of a model in the early stages of training and searching for points where gradients align between domains. By discovering a set of parameter values where gradients are updated towards the same direction for each data distribution present in the training set, the proposed Gradient-Guided Annealing (GGA) algorithm encourages models to seek out minima that exhibit improved robustness against domain shifts. The efficacy of GGA is evaluated on five widely accepted and challenging image classification domain generalization benchmarks, where its use alone is able to establish highly competitive or even state-of-the-art performance. Moreover, when combined with previously proposed domain-generalization algorithms it is able to consistently improve their effectiveness by significant margins.
Detection and classification of pulmonary nodules is a challenge in medical image analysis due to the variety of shapes and sizes of nodules and their high concealment. Despite the success of traditional deep learning methods in image classification, deep networks still struggle to perfectly capture subtle changes in lung nodule detection. Therefore, we propose a residual multi-task network (Res-MTNet) model, which combines multi-task learning and residual learning, and improves feature representation ability by sharing feature extraction layer and introducing residual connections. Multi-task learning enables the model to handle multiple tasks simultaneously, while the residual module solves the problem of disappearing gradients, ensuring stable training of deeper networks and facilitating information sharing between tasks. Res-MTNet enhances the robustness and accuracy of the model, providing a more reliable lung nodule analysis tool for clinical medicine and telemedicine.
Understanding the classifications of deep neural networks, e.g. used in safety-critical situations, is becoming increasingly important. While recent models can locally explain a single decision, to provide a faithful global explanation about an accurate model's general behavior is a more challenging open task. Towards that goal, we introduce the Quadratic Programming Enhanced Model (QPM), which learns globally interpretable class representations. QPM represents every class with a binary assignment of very few, typically 5, features, that are also assigned to other classes, ensuring easily comparable contrastive class representations. This compact binary assignment is found using discrete optimization based on predefined similarity measures and interpretability constraints. The resulting optimal assignment is used to fine-tune the diverse features, so that each of them becomes the shared general concept between the assigned classes. Extensive evaluations show that QPM delivers unprecedented global interpretability across small and large-scale datasets while setting the state of the art for the accuracy of interpretable models.
Although data augmentation is an effective method to address the imbalanced-small sample data (ISSD) problem in hyperspectral image classification (HSIC), most methodologies extend features in the latent space. Few, however, generate realistic and diverse samples using text information to balance the limited number of annotated samples. Recently, text-driven diffusion models have gained significant attention due to their remarkable ability to generate highly diverse images based on given text prompts in natural image synthesis. Therefore, this paper proposes a novel language-informed hyperspectral image synthesis method (Txt2HSI-LDM(VAE)) for addressing the ISSD problem of HSIC. First, for addressing the high-dimensional hyperspectral data, we use universal varitional autoencoeder (VAE) to map the hyperspectral into a low-dimensional latent space and get stable feature representation, which hugely reduce the inference parameter of diffusion model. Next, a semi-supervised diffusion model is designed for fully taking advantage of unlabeled data, beside, random polygon spatial clipping (RPSC) and uncertainty estimation of latent feature (LF-UE) are also used for simulating the varying degrees of mixing of training data. Then, VAE decodes HSI from latent space generated by diffusion model with the conditional language as input, contributing to more realistic and diverse samples. In our experiments, we fully evaluate the effectiveness of synthetic samples from aspect of statistical characteristic and data distribution in 2D-PCA space. Additionally, cross-attention map is visualized on the pixel-level to prove that our proposed model can capture the spatial layout of and geometry of the generated hyperspectral image depend on the visual-linguistic alignment.
Recent advancements in deep learning for image classification predominantly rely on convolutional neural networks (CNNs) or Transformer-based architectures. However, these models face notable challenges in medical imaging, particularly in capturing intricate texture details and contextual features. Kolmogorov-Arnold Networks (KANs) represent a novel class of architectures that enhance nonlinear transformation modeling, offering improved representation of complex features. In this work, we present MedKAN, a medical image classification framework built upon KAN and its convolutional extensions. MedKAN features two core modules: the Local Information KAN (LIK) module for fine-grained feature extraction and the Global Information KAN (GIK) module for global context integration. By combining these modules, MedKAN achieves robust feature modeling and fusion. To address diverse computational needs, we introduce three scalable variants--MedKAN-S, MedKAN-B, and MedKAN-L. Experimental results on nine public medical imaging datasets demonstrate that MedKAN achieves superior performance compared to CNN- and Transformer-based models, highlighting its effectiveness and generalizability in medical image analysis.
The recognition of art styles and authors is crucial in areas like cultural heritage protection, art market analysis, and historical research. With the advancement of deep learning, Convolutional Neural Networks (CNNs) and Transformer models have become key tools for image classification. While CNNs excel in local feature extraction, they struggle with global context, and Transformers are strong in capturing global dependencies but weak in fine-grained local details. To address these challenges, this paper proposes a fusion model combining CNNs and Transformers for art style and author recognition. The model first extracts local features using CNNs, then captures global context with a Transformer, followed by a feature fusion mechanism to enhance classification accuracy. Experiments on Chinese and oil painting datasets show the fusion model outperforms individual CNN and Transformer models, improving classification accuracy by 9.7% and 7.1%, respectively, and increasing F1 scores by 0.06 and 0.05. The results demonstrate the model's effectiveness and potential for future improvements, such as multimodal integration and architecture optimization.
Due to the difficulty of obtaining labeled data for hyperspectral images (HSIs), cross-scene classification has emerged as a widely adopted approach in the remote sensing community. It involves training a model using labeled data from a source domain (SD) and unlabeled data from a target domain (TD), followed by inferencing on the TD. However, variations in the reflectance spectrum of the same object between the SD and the TD, as well as differences in the feature distribution of the same land cover class, pose significant challenges to the performance of cross-scene classification. To address this issue, we propose a dual classification head self-training network (DHSNet). This method aligns class-wise features across domains, ensuring that the trained classifier can accurately classify TD data of different classes. We introduce a dual classification head self-training strategy for the first time in the cross-scene HSI classification field. The proposed approach mitigates domain gap while preventing the accumulation of incorrect pseudo-labels in the model. Additionally, we incorporate a novel central feature attention mechanism to enhance the model's capacity to learn scene-invariant features across domains. Experimental results on three cross-scene HSI datasets demonstrate that the proposed DHSNET significantly outperforms other state-of-the-art approaches. The code for DHSNet will be available at https://github.com/liurongwhm.
The remarkable success of deep learning in recent years has prompted applications in medical image classification and diagnosis tasks. While classification models have demonstrated robustness in classifying simpler datasets like MNIST or natural images such as ImageNet, this resilience is not consistently observed in complex medical image datasets where data is more scarce and lacks diversity. Moreover, previous findings on natural image datasets have indicated a potential trade-off between data likelihood and classification accuracy. In this study, we explore the use of score-based generative models as classifiers for medical images, specifically mammographic images. Our findings suggest that our proposed generative classifier model not only achieves superior classification results on CBIS-DDSM, INbreast and Vin-Dr Mammo datasets, but also introduces a novel approach to image classification in a broader context. Our code is publicly available at https://github.com/sushmitasarker/sgc_for_medical_image_classification
We formulate truncated singular value decompositions of entire convolutional neural networks. We demonstrate the computed left and right singular vectors are useful in identifying which images the convolutional neural network is likely to perform poorly on. To create this diagnostic tool, we define two metrics: the Right Projection Ratio and the Left Projection Ratio. The Right (Left) Projection Ratio evaluates the fidelity of the projection of an image (label) onto the computed right (left) singular vectors. We observe that both ratios are able to identify the presence of class imbalance for an image classification problem. Additionally, the Right Projection Ratio, which only requires unlabeled data, is found to be correlated to the model's performance when applied to image segmentation. This suggests the Right Projection Ratio could be a useful metric to estimate how likely the model is to perform well on a sample.
Transformers have been favored in both uni-modal and multi-modal foundation models for their flexible scalability in attention modules. Consequently, a number of pre-trained Transformer models, e.g., LLaVA, CLIP, and DEIT, are publicly available. Recent research has introduced subquadratic architectures like Mamba, which enables global awareness with linear complexity. Nevertheless, training specialized subquadratic architectures from scratch for certain tasks is both resource-intensive and time-consuming. As a motivator, we explore cross-architecture training to transfer the ready knowledge in existing Transformer models to alternative architecture Mamba, termed TransMamba. Our approach employs a two-stage strategy to expedite training new Mamba models, ensuring effectiveness in across uni-modal and cross-modal tasks. Concerning architecture disparities, we project the intermediate features into an aligned latent space before transferring knowledge. On top of that, a Weight Subcloning and Adaptive Bidirectional distillation method (WSAB) is introduced for knowledge transfer without limitations on varying layer counts. For cross-modal learning, we propose a cross-Mamba module that integrates language awareness into Mamba's visual features, enhancing the cross-modal interaction capabilities of Mamba architecture. Despite using less than 75% of the training data typically required for training from scratch, TransMamba boasts substantially stronger performance across various network architectures and downstream tasks, including image classification, visual question answering, and text-video retrieval. The code will be publicly available.
Transformer models typically calculate attention matrices using dot products, which have limitations when capturing nonlinear relationships between embedding vectors. We propose Neural Attention, a technique that replaces dot products with feed-forward networks, enabling a more expressive representation of relationships between tokens. This approach modifies only the attention matrix calculation while preserving the matrix dimensions, making it easily adaptable to existing transformer-based architectures. We provide a detailed mathematical justification for why Neural Attention increases representational capacity and conduct controlled experiments to validate this claim. When comparing Neural Attention and Dot-Product Attention, NLP experiments on WikiText-103 show a reduction in perplexity of over 5 percent. Similarly, experiments on CIFAR-10 and CIFAR-100 show comparable improvements for image classification tasks. While Neural Attention introduces higher computational demands, we develop techniques to mitigate these challenges, ensuring practical usability without sacrificing the increased expressivity it provides. This work establishes Neural Attention as an effective means of enhancing the predictive capabilities of transformer models across a variety of applications.