With the rapid advancement and increased use of deep learning models in image identification, security becomes a major concern to their deployment in safety-critical systems. Since the accuracy and robustness of deep learning models are primarily attributed from the purity of the training samples, therefore the deep learning architectures are often susceptible to adversarial attacks. Adversarial attacks are often obtained by making subtle perturbations to normal images, which are mostly imperceptible to humans, but can seriously confuse the state-of-the-art machine learning models. What is so special in the slightest intelligent perturbations or noise additions over normal images that it leads to catastrophic classifications by the deep neural networks? Using statistical hypothesis testing, we find that Conditional Variational AutoEncoders (CVAE) are surprisingly good at detecting imperceptible image perturbations. In this paper, we show how CVAEs can be effectively used to detect adversarial attacks on image classification networks. We demonstrate our results over MNIST, CIFAR-10 dataset and show how our method gives comparable performance to the state-of-the-art methods in detecting adversaries while not getting confused with noisy images, where most of the existing methods falter.
Single-domain generalization aims to learn a model from single source domain data to achieve generalized performance on other unseen target domains. Existing works primarily focus on improving the generalization ability of static networks. However, static networks are unable to dynamically adapt to the diverse variations in different image scenes, leading to limited generalization capability. Different scenes exhibit varying levels of complexity, and the complexity of images further varies significantly in cross-domain scenarios. In this paper, we propose a dynamic object-centric perception network based on prompt learning, aiming to adapt to the variations in image complexity. Specifically, we propose an object-centric gating module based on prompt learning to focus attention on the object-centric features guided by the various scene prompts. Then, with the object-centric gating masks, the dynamic selective module dynamically selects highly correlated feature regions in both spatial and channel dimensions enabling the model to adaptively perceive object-centric relevant features, thereby enhancing the generalization capability. Extensive experiments were conducted on single-domain generalization tasks in image classification and object detection. The experimental results demonstrate that our approach outperforms state-of-the-art methods, which validates the effectiveness and generally of our proposed method.
All-in-one image restoration, addressing diverse degradation types with a unified model, presents significant challenges in designing task-specific prompts that effectively guide restoration across multiple degradation scenarios. While adaptive prompt learning enables end-to-end optimization, it often yields overlapping or redundant task representations. Conversely, explicit prompts derived from pretrained classifiers enhance discriminability but may discard critical visual information for reconstruction. To address these limitations, we introduce Contrastive Prompt Learning (CPL), a novel framework that fundamentally enhances prompt-task alignment through two complementary innovations: a \emph{Sparse Prompt Module (SPM)} that efficiently captures degradation-specific features while minimizing redundancy, and a \emph{Contrastive Prompt Regularization (CPR)} that explicitly strengthens task boundaries by incorporating negative prompt samples across different degradation types. Unlike previous approaches that focus primarily on degradation classification, CPL optimizes the critical interaction between prompts and the restoration model itself. Extensive experiments across five comprehensive benchmarks demonstrate that CPL consistently enhances state-of-the-art all-in-one restoration models, achieving significant improvements in both standard multi-task scenarios and challenging composite degradation settings. Our framework establishes new state-of-the-art performance while maintaining parameter efficiency, offering a principled solution for unified image restoration.
We introduce II-20 (Image Insight 2020), a multimedia analytics approach for analytic categorization of image collections. Advanced visualizations for image collections exist, but they need tight integration with a machine model to support analytic categorization. Directly employing computer vision and interactive learning techniques gravitates towards search. Analytic categorization, however, is not machine classification (the difference between the two is called the pragmatic gap): a human adds/redefines/deletes categories of relevance on the fly to build insight, whereas the machine classifier is rigid and non-adaptive. Analytic categorization that brings the user to insight requires a flexible machine model that allows dynamic sliding on the exploration-search axis, as well as semantic interactions. II-20 brings 3 major contributions to multimedia analytics on image collections and towards closing the pragmatic gap. Firstly, a machine model that closely follows the user's interactions and dynamically models her categories of relevance. II-20's model, in addition to matching and exceeding the state of the art w. r. t. relevance, allows the user to dynamically slide on the exploration-search axis without additional input from her side. Secondly, the dynamic, 1-image-at-a-time Tetris metaphor that synergizes with the model. It allows the model to analyze the collection by itself with minimal interaction from the user and complements the classic grid metaphor. Thirdly, the fast-forward interaction, allowing the user to harness the model to quickly expand ("fast-forward") the categories of relevance, expands the multimedia analytics semantic interaction dictionary. Automated experiments show that II-20's model outperforms the state of the art and also demonstrate Tetris's analytic quality. User studies confirm that II-20 is an intuitive, efficient, and effective multimedia analytics tool.
The traditional object retrieval task aims to learn a discriminative feature representation with intra-similarity and inter-dissimilarity, which supposes that the objects in an image are manually or automatically pre-cropped exactly. However, in many real-world searching scenarios (e.g., video surveillance), the objects (e.g., persons, vehicles, etc.) are seldom accurately detected or annotated. Therefore, object-level retrieval becomes intractable without bounding-box annotation, which leads to a new but challenging topic, i.e. image-level search. In this paper, to address the image search issue, we first introduce an end-to-end Integrated Net (I-Net), which has three merits: 1) A Siamese architecture and an on-line pairing strategy for similar and dissimilar objects in the given images are designed. 2) A novel on-line pairing (OLP) loss is introduced with a dynamic feature dictionary, which alleviates the multi-task training stagnation problem, by automatically generating a number of negative pairs to restrict the positives. 3) A hard example priority (HEP) based softmax loss is proposed to improve the robustness of classification task by selecting hard categories. With the philosophy of divide and conquer, we further propose an improved I-Net, called DC-I-Net, which makes two new contributions: 1) two modules are tailored to handle different tasks separately in the integrated framework, such that the task specification is guaranteed. 2) A class-center guided HEP loss (C2HEP) by exploiting the stored class centers is proposed, such that the intra-similarity and inter-dissimilarity can be captured for ultimate retrieval. Extensive experiments on famous image-level search oriented benchmark datasets demonstrate that the proposed DC-I-Net outperforms the state-of-the-art tasks-integrated and tasks-separated image search models.
Medical head CT-scan imaging has been successfully combined with deep learning for medical diagnostics of head diseases and lesions[1]. State of the art classification models and algorithms for this task usually are based on 3d convolution layers for volumetric data on a supervised learning setting (1 input volume, 1 prediction per patient) or 2d convolution layers in a supervised setting (1 input image, 1 prediction per image). However a very common scenario in developing countries is to have the volume metadata lost due multiple reasons for example formatting conversion in images (for example .dicom to jpg), in this scenario the doctor analyses the collection of images and then emits a single diagnostic for the patient (with possibly an unfixed and variable number of images per patient) , this prevents it from being possible to use state of the art 3d models, but also is not possible to convert it to a supervised problem in a (1 image,1 diagnostic) setting because different angles or positions of the images for a single patient may not contain the disease or lesion. In this study we propose a solution for this scenario by combining 2d convolutional[2] models with sequence models which generate a prediction only after all images have been processed by the model for a given patient \(i\), this creates a multi-image to single-diagnostic setting \(y^i=f(x_1,x_2,..,x_n)\) where \(n\) may be different between patients. The experimental results demonstrate that it is possible to get a multi-image to single diagnostic model which mimics human doctor diagnostic process: evaluate the collection of patient images and then use important information in memory to decide a single diagnostic for the patient.
Masked Image Modeling (MIM) has recently been established as a potent pre-training paradigm. A pretext task is constructed by masking patches in an input image, and this masked content is then predicted by a neural network using visible patches as sole input. This pre-training leads to state-of-the-art performance when finetuned for high-level semantic tasks, e.g. image classification and object detection. In this paper we instead seek to learn representations that transfer well to a wide variety of 3D vision and lower-level geometric downstream tasks, such as depth prediction or optical flow estimation. Inspired by MIM, we propose an unsupervised representation learning task trained from pairs of images showing the same scene from different viewpoints. More precisely, we propose the pretext task of cross-view completion where the first input image is partially masked, and this masked content has to be reconstructed from the visible content and the second image. In single-view MIM, the masked content often cannot be inferred precisely from the visible portion only, so the model learns to act as a prior influenced by high-level semantics. In contrast, this ambiguity can be resolved with cross-view completion from the second unmasked image, on the condition that the model is able to understand the spatial relationship between the two images. Our experiments show that our pretext task leads to significantly improved performance for monocular 3D vision downstream tasks such as depth estimation. In addition, our model can be directly applied to binocular downstream tasks like optical flow or relative camera pose estimation, for which we obtain competitive results without bells and whistles, i.e., using a generic architecture without any task-specific design.
In this work we introduce a novel medical image style transfer method, StyleMapper, that can transfer medical scans to an unseen style with access to limited training data. This is made possible by training our model on unlimited possibilities of simulated random medical imaging styles on the training set, making our work more computationally efficient when compared with other style transfer methods. Moreover, our method enables arbitrary style transfer: transferring images to styles unseen in training. This is useful for medical imaging, where images are acquired using different protocols and different scanner models, resulting in a variety of styles that data may need to be transferred between. Methods: Our model disentangles image content from style and can modify an image's style by simply replacing the style encoding with one extracted from a single image of the target style, with no additional optimization required. This also allows the model to distinguish between different styles of images, including among those that were unseen in training. We propose a formal description of the proposed model. Results: Experimental results on breast magnetic resonance images indicate the effectiveness of our method for style transfer. Conclusion: Our style transfer method allows for the alignment of medical images taken with different scanners into a single unified style dataset, allowing for the training of other downstream tasks on such a dataset for tasks such as classification, object detection and others.
Self-training is an important class of unsupervised domain adaptation (UDA) approaches that are used to mitigate the problem of domain shift, when applying knowledge learned from a labeled source domain to unlabeled and heterogeneous target domains. While self-training-based UDA has shown considerable promise on discriminative tasks, including classification and segmentation, through reliable pseudo-label filtering based on the maximum softmax probability, there is a paucity of prior work on self-training-based UDA for generative tasks, including image modality translation. To fill this gap, in this work, we seek to develop a generative self-training (GST) framework for domain adaptive image translation with continuous value prediction and regression objectives. Specifically, we quantify both aleatoric and epistemic uncertainties within our GST using variational Bayes learning to measure the reliability of synthesized data. We also introduce a self-attention scheme that de-emphasizes the background region to prevent it from dominating the training process. The adaptation is then carried out by an alternating optimization scheme with target domain supervision that focuses attention on the regions with reliable pseudo-labels. We evaluated our framework on two cross-scanner/center, inter-subject translation tasks, including tagged-to-cine magnetic resonance (MR) image translation and T1-weighted MR-to-fractional anisotropy translation. Extensive validations with unpaired target domain data showed that our GST yielded superior synthesis performance in comparison to adversarial training UDA methods.
Recent advancements in cognitive computing, with the integration of deep learning techniques, have facilitated the development of intelligent cognitive systems (ICS). This is particularly beneficial in the context of rail defect detection, where the ICS would emulate human-like analysis of image data for defect patterns. Despite the success of Convolutional Neural Networks (CNN) in visual defect classification, the scarcity of large datasets for rail defect detection remains a challenge due to infrequent accident events that would result in defective parts and images. Contemporary researchers have addressed this data scarcity challenge by exploring rule-based and generative data augmentation models. Among these, Variational Autoencoder (VAE) models can generate realistic data without extensive baseline datasets for noise modeling. This study proposes a VAE-based synthetic image generation technique for rail defects, incorporating weight decay regularization and image reconstruction loss to prevent overfitting. The proposed method is applied to create a synthetic dataset for the Canadian Pacific Railway (CPR) with just 50 real samples across five classes. Remarkably, 500 synthetic samples are generated with a minimal reconstruction loss of 0.021. A Visual Transformer (ViT) model underwent fine-tuning using this synthetic CPR dataset, achieving high accuracy rates (98%-99%) in classifying the five defect classes. This research offers a promising solution to the data scarcity challenge in rail defect detection, showcasing the potential for robust ICS development in this domain.
We describe a new class of subsampling techniques for CNNs, termed multisampling, that significantly increases the amount of information kept by feature maps through subsampling layers. One version of our method, which we call checkered subsampling, significantly improves the accuracy of state-of-the-art architectures such as DenseNet and ResNet without any additional parameters and, remarkably, improves the accuracy of certain pretrained ImageNet models without any training or fine-tuning. We glean possible insight into the nature of data augmentations and demonstrate experimentally that coarse feature maps are bottlenecking the performance of neural networks in image classification.
The wide adoption of Convolutional Neural Networks (CNNs) in applications where decision-making under uncertainty is fundamental, has brought a great deal of attention to the ability of these models to accurately quantify the uncertainty in their predictions. Previous work on combining CNNs with Gaussian processes (GPs) has been developed under the assumption that the predictive probabilities of these models are well-calibrated. In this paper we show that, in fact, current combinations of CNNs and GPs are miscalibrated. We proposes a novel combination that considerably outperforms previous approaches on this aspect, while achieving state-of-the-art performance on image classification tasks.
Deep learning has proven itself as a successful set of models for learning useful semantic representations of data. These, however, are mostly implicitly learned as part of a classification task. In this paper we propose the triplet network model, which aims to learn useful representations by distance comparisons. A similar model was defined by Wang et al. (2014), tailor made for learning a ranking for image information retrieval. Here we demonstrate using various datasets that our model learns a better representation than that of its immediate competitor, the Siamese network. We also discuss future possible usage as a framework for unsupervised learning.
In this paper we provide an approach for deep learning that protects against adversarial examples in image classification-type networks. The approach relies on two mechanisms:1) a mechanism that increases robustness at the expense of accuracy, and, 2) a mechanism that improves accuracy but does not always increase robustness. We show that an approach combining the two mechanisms can provide protection against adversarial examples while retaining accuracy. We formulate potential attacks on our approach with experimental results to demonstrate its effectiveness. We also provide a robustness guarantee for our approach along with an interpretation for the guarantee.
In the proposed SEHybridSN model, a dense block was used to reuse shallow feature and aimed at better exploiting hierarchical spatial spectral feature. Subsequent depth separable convolutional layers were used to discriminate the spatial information. Further refinement of spatial spectral features was realized by the channel attention method, which were performed behind every 3D convolutional layer and every 2D convolutional layer. Experiment results indicate that our proposed model learn more discriminative spatial spectral features using very few training data. SEHybridSN using only 0.05 and 0.01 labeled data for training, a very satisfactory performance is obtained.
Image classification is a significant challenge in computer vision, particularly in domains humans are not accustomed to. As machine learning and artificial intelligence become more prominent, it is crucial these algorithms develop a sense of sight that is on par with or exceeds human ability. For this reason, we have collected, cleaned, and parsed a large dataset of hand-drawn doodles and compared multiple machine learning solutions to classify these images into 170 distinct categories. The best model we found achieved a Top-1 accuracy of 47.5%, significantly surpassing human performance on the dataset, which stands at 41%.
Modern neural networks (NNs) often do not generalize well in the presence of a "covariate shift"; that is, in situations where the training and test data distributions differ, but the conditional distribution of classification labels remains unchanged. In such cases, NN generalization can be reduced to a problem of learning more domain-invariant features. Domain adaptation (DA) methods include a range of techniques aimed at achieving this; however, these methods have struggled with the need for extensive hyperparameter tuning, which then incurs significant computational costs. In this work, we introduce SIDDA, an out-of-the-box DA training algorithm built upon the Sinkhorn divergence, that can achieve effective domain alignment with minimal hyperparameter tuning and computational overhead. We demonstrate the efficacy of our method on multiple simulated and real datasets of varying complexity, including simple shapes, handwritten digits, and real astronomical observations. SIDDA is compatible with a variety of NN architectures, and it works particularly well in improving classification accuracy and model calibration when paired with equivariant neural networks (ENNs). We find that SIDDA enhances the generalization capabilities of NNs, achieving up to a $\approx40\%$ improvement in classification accuracy on unlabeled target data. We also study the efficacy of DA on ENNs with respect to the varying group orders of the dihedral group $D_N$, and find that the model performance improves as the degree of equivariance increases. Finally, we find that SIDDA enhances model calibration on both source and target data--achieving over an order of magnitude improvement in the ECE and Brier score. SIDDA's versatility, combined with its automated approach to domain alignment, has the potential to advance multi-dataset studies by enabling the development of highly generalizable models.
Convolutional Neural Networks (CNNs) specialize in feature extraction rather than function mapping. In doing so they form complex internal hierarchical feature representations, the complexity of which gradually increases with a corresponding increment in neural network depth. In this paper, we examine the feature extraction capabilities of CNNs using Maximum Entropy (ME) and Signal-to-Noise Ratio (SNR) to validate the idea that, CNN models should be tailored for a given task and complexity of the input data. SNR and ME measures are used as they can accurately determine in the input dataset, the relative amount of signal information to the random noise and the maximum amount of information respectively. We use two well known benchmarking datasets, MNIST and CIFAR-10 to examine the information extraction and abstraction capabilities of CNNs. Through our experiments, we examine convolutional feature extraction and abstraction capabilities in CNNs and show that the classification accuracy or performance of CNNs is greatly dependent on the amount, complexity and quality of the signal information present in the input data. Furthermore, we show the effect of information overflow and underflow on CNN classification accuracies. Our hypothesis is that the feature extraction and abstraction capabilities of convolutional layers are limited and therefore, CNN models should be tailored to the input data by using appropriately sized CNNs based on the SNR and ME measures of the input dataset.
This paper presents a new model architecture, gate with inhibition MLP (giMLP).The gate with inhibition on CycleMLP (gi-CycleMLP) can produce equal performance on the ImageNet classification task, and it also improves the BERT, Roberta, and DeBERTaV3 models depending on two novel techniques. The first is the gating MLP, where matrix multiplications between the MLP and the trunk Attention input in further adjust models' adaptation. The second is inhibition which inhibits or enhances the branch adjustment, and with the inhibition levels increasing, it offers models more muscular features restriction. We show that the giCycleMLP with a lower inhibition level can be competitive with the original CycleMLP in terms of ImageNet classification accuracy. In addition, we also show through a comprehensive empirical study that these techniques significantly improve the performance of fine-tuning NLU downstream tasks. As for the gate with inhibition MLPs on DeBERTa (giDeBERTa) fine-tuning, we find it can achieve appealing results on most parts of NLU tasks without any extra pretraining again. We also find that with the use of Gate With Inhibition, the activation function should have a short and smooth negative tail, with which the unimportant features or the features that hurt models can be moderately inhibited. The experiments on ImageNet and twelve language downstream tasks demonstrate the effectiveness of Gate With Inhibition, both for image classification and for enhancing the capacity of nature language fine-tuning without any extra pretraining.
Despite the extensive body of literature focused on remote sensing applications for land cover mapping and the availability of high-resolution satellite imagery, methods for continuously updating classification maps in real-time remain limited, especially when training data is scarce. This paper introduces the Recursive Bayesian Classifier (RBC), which converts any instantaneous classifier into a robust online method through a probabilistic framework that is resilient to non-informative image variations. Three experiments are conducted using Sentinel-2 data: water mapping of the Oroville Dam in California and the Charles River basin in Massachusetts, and deforestation detection in the Amazon. RBC is applied to a Gaussian Mixture Model (GMM), Logistic Regression (LR), and our proposed Spectral Index Classifier (SIC). Results show that RBC significantly enhances classifier robustness in multitemporal settings under challenging conditions, such as cloud cover and cyanobacterial blooms. Specifically, balanced classification accuracy improves by up to 26.95% for SIC, 12.4% for GMM, and 13.81% for LR in water mapping, and by 15.25%, 14.17%, and 14.7% in deforestation detection. Moreover, without additional training data, RBC improves the performance of the state-of-the-art DeepWaterMap and WatNet algorithms by up to 9.62% and 11.03%. These benefits are provided by RBC while requiring minimal supervision and maintaining a low computational cost that remains constant for each time step regardless of the time-series length.