Effective image classification in real-world scenarios requires models capable of capturing both local features and global context, while adapting to the evolving complexity of the input. This paper presents a hybrid deep learning system named AMR-Net (Adaptive Modular Retrieval Network) for adaptable image classification in environmental and industrial contexts. The proposed model integrates ResNet-based local feature extraction and Vision Transformer-based global representation learning with an FAISS-based memory retrieval module for contextual enhancement. Furthermore, AMR-Net has an adaptive modular extension mechanism to dynamically enhance network capacity when validation performance plateaus during training. The proposed framework was evaluated using a ten-class waste classification dataset and the Kolektor Surface Defect Dataset 2 (KolektorSDD2), achieving classification accuracies of 97.15% and 99.80%, respectively. Experimental results demonstrate that AMR-Net consistently outperforms mainstream convolutional and transformer-based approaches while maintaining robust cross-domain generalization. Component-wise ablation studies show that global attention (ViT), retrieval-based contextual learning (FAISS), and adaptive modular growth each provide complementary and partially overlapping contributions to overall performance, with global attention contributing the largest individual gain. The proposed architecture is a scalable and flexible solution for image classification challenges characterized by data unpredictability, class imbalance, and dynamic feature distributions.
Three-dimensional medical image data and computer-aided decision making, particularly using deep learning, are becoming increasingly important in the medical field. To aid in these developments we introduce PR3DICTR: Platform for Research in 3D Image Classification and sTandardised tRaining. Built using community-standard distributions (PyTorch and MONAI), PR3DICTR provides an open-access, flexible and convenient framework for prediction model development, with an explicit focus on classification using three-dimensional medical image data. By combining modular design principles and standardization, it aims to alleviate developmental burden whilst retaining adjustability. It provides users with a wealth of pre-established functionality, for instance in model architecture design options, hyper-parameter solutions and training methodologies, but still gives users the opportunity and freedom to ``plug in'' their own solutions or modules. PR3DICTR can be applied to any binary or event-based three-dimensional classification task and can work with as little as two lines of code.
We propose a test-time defense mechanism against adversarial attacks: imperceptible image perturbations that significantly alter the predictions of a model. Unlike existing methods that rely on feature filtering or smoothing, which can lead to information loss, we propose to "combat noise with noise" by leveraging stochastic resonance to enhance robustness while minimizing information loss. Our approach introduces small translational perturbations to the input image, aligns the transformed feature embeddings, and aggregates them before mapping back to the original reference image. This can be expressed in a closed-form formula, which can be deployed on diverse existing network architectures without introducing additional network modules or fine-tuning for specific attack types. The resulting method is entirely training-free, architecture-agnostic, and attack-agnostic. Empirical results show state-of-the-art robustness on image classification and, for the first time, establish a generic test-time defense for dense prediction tasks, including stereo matching and optical flow, highlighting the method's versatility and practicality. Specifically, relative to clean (unperturbed) performance, our method recovers up to 68.1% of the accuracy loss on image classification, 71.9% on stereo matching, and 29.2% on optical flow under various types of adversarial attacks.
Concept Bottleneck Models (CBMs) enhance the interpretability of deep neural networks by aligning the features extracted from images with human-interpretable concepts. However, conventional CBMs typically use a linear classifier that maps concept activations to a fixed set of predefined classes, limiting their flexibility in open-vocabulary settings. In this paper, we propose Multimodal Concept Bottleneck Model (MM-CBM) to address these issues and extend CBMs into CLIP structure. MM-CBM utilizes dual Concept Bottleneck Layers (CBLs) to align both the image and text embeddings into interpretable features. This allows us to perform new vision tasks like zero-shot classification or image retrieval in an interpretable way. Compared to existing methods, MM-CBM achieves 23.28% accuracy improvement on average across four standard benchmarks. Our method maintains competitive performance compared with black-box CLIP model while offering interpretability. Under classification tasks, MM-CBM's accuracy is on average 3.94% lower than CLIP in the target-dataset adaptation setting and 9.84% in the zero-shot transfer setting. Under image-text retrieval tasks, it achieves an average gap of 2.48% for image retrieval and 9.22% for text retrieval.
Deep joint source--channel coding (Deep JSCC) can suffer from domain shift between training and deployment data. We study single-source domain adaptation for image classification over wireless channels. A Gaussian example shows that improving channel quality does not necessarily improve target-domain accuracy under a fixed source-trained classifier. We further proposed an information objective to reduce domain information while preserving class information. Motivated by the analysis, we develop a domain-adaptive Deep JSCC method that combines class-wise adversarial alignment with pseudo-label supervised contrastive learning. The former uses all target samples to reduce class-conditional domain differences, while the latter selects high-confidence target samples to obtain class-discriminative received representations. We relate these losses to the information objective, and analyze the effects of pseudo-label errors and confidence-based selection. Experiments on SVHN--MNIST and PACS over AWGN and Rayleigh channels demonstrate improved target-domain classification without additional inference-time networks. The proposed method achieves 98.15\% accuracy on SVHN $\rightarrow$ MNIST at 10~dB CSNR.
Prototypes in classification models are almost always vectors, and a vector has no readable form. This paper asks what happens when a prototype is an image. Documents give the question a natural form, because a document can be rendered as a multi-channel image in which every token becomes a pixel, so a class representative can take the same shape and the same channel semantics as the inputs it stands for. ProtoSemImage represents each class by one or more visual archetypes: prototype images in a four-channel HSV space whose channels carry named linguistic factors. A Skip-Gram objective learns that color space end to end through a four-dimensional bottleneck, discourse boundary rows become differentiable typed difference rows, and classification reduces to 2D visual template matching: a deformable row alignment between a document image and the archetype bank, in the spirit of dynamic time warping. Because the match is a spatial pattern comparison rather than a linear readout, the model reports where an input departs from its archetype and along which channel, and a generative head decodes each archetype back into text. The image representation works: it beats an otherwise identical model with vector prototypes in all three paired seeds, by between 4.3 and 11.8 points on a ten-class task. The distance-based matching does not. A diagnostic that keeps the representation fixed and swaps only the classifier recovers the sequence baselines, which locates a 20.6-point shortfall in the matching rather than in the color compression, and a benchmark built so that a pair of documents shares a bag of words and differs only in arrangement confirms the layout-preservation it was designed for. We report both directions, because for a representation whose whole purpose is inspect ability, the failure modes are as informative as the gains.
Lesion segmentation and classification in breast ultrasound inform each other, yet many multi-task networks couple the two tasks only through shared features. This paper introduces a decoder in which a Task Interaction Module (TIM) exchanges latent features between the segmentation and classification branches at each of four levels, and an Adaptive Interaction Weighting (AIW) unit uses activation dispersion to set, per image and level, one softmax budget that the two directions share. Multi-scale context fusion (MSCF) adds dilated context to the residual blocks. On BUSI and BUSI-WHU, split so that duplicate images and, for BUSI-WHU, images of one patient stay within one partition, the model exceeds the mean IoU of CNN, transformer, multi-task and same-backbone baselines, reaching 72.9% and 85.4%, and flags lesion pixels in 19.0% of its predictions on normal BUSI test images against 33.3% for MTI-Net. Replacing the per-image budget by fixed or learned constant coefficients, or by independent sigmoid gates, costs 0.92 to 1.51 IoU points on BUSI. Code and data partitions: https://github.com/C-loud-Nine/Adaptive-Task-Interaction-BUS
Few-shot image classification (FSIC) frequently suffers from visual ambiguity, where background clutter and RGB appearance similarities hinder discriminative feature matching in data-scarce settings. To mitigate this, we propose DAFIC, a geometry-guided framework that incorporates monocular depth into representation learning, episodic adaptation, and dense support-query matching. DAFIC first learns geometry-aware token representations through RGB-D self-supervised pre-training. Subsequently, the Depth-guided Geometric Adaptation (DGA) module uses depth-derived geometry to refine patch representations with geometry-enriched class context while suppressing background interference. Spatial Bidirectional Patch Matching (SBiM) then performs bidirectional dense matching on the refined features, reducing sensitivity to spurious correspondences and background clutter. Extensive experiments on four benchmarks show that DAFIC achieves strong performance against existing methods, with ablation and robustness analyses supporting the effectiveness of coupling geometry-conditioned episodic adaptation with bidirectional matching for few-shot visual recognition.
We study Joint Flow Matching (JFM), a flow-based framework for learning a joint distribution over paired variables such as images and class labels. JFM transports the variables in opposite directions: an image is transported from noise to data while its label is transported from data to noise. This provides a single probabilistic model in which generation and classification are represented as complementary conditional inferences. We formulate the joint distribution induced by this construction and characterise the relationship between its forward and reverse conditionals. We investigate the resulting model for classification and conditional generation, including confidence calibration and consistency between generated samples and classifier predictions. Experiments on conditional datasets demonstrate competitive classification accuracy, well-calibrated confidence without post-hoc calibration, and classifier-consistent generation. We further explore the shared joint distribution as a basis for interpreting the relationship between discriminative and generative behaviour.
Vision-language models (VLMs) show promise for large-scale affect inference from images, but their alignment with human affective ratings remains unclear. We benchmarked 16 proprietary and open-source VLMs on three validated affective image datasets: the International Affective Picture System, the Nencki Affective Picture System, and the Library of AI-Generated Affective Images. In a zero-shot setting, models performed two tasks: (i) top-emotion classification and (ii) prediction of human ratings on 1--7/9 Likert scales for discrete emotions and affective dimensions. We also evaluated rater-conditioned prompting on the LAI-GAI dataset using de-identified participant metadata. Best models showed strong discrete emotion classification (60%--80% on six-emotion tasks; 60%--75% on a 12-category task), with anger and surprise predicted least accurately. For continuous ratings, models aligned moderately to strongly with human judgments ($r > 0.75$) but showed consistent biases, especially weaker performance on arousal and overestimation of response intensity. Rater-conditioned prompting had only small, inconsistent effects. Overall, VLMs captured broad affective patterns but sometimes lacked the nuance, although newer model generations performed better.
Vision-language models (VLMs) for radiology have emerged as a scalable paradigm by leveraging image-report pairs naturally produced in clinical workflows. However, this pairing reveals a mismatch in scale: each finding occupies only a small region of the image, yet supervision is provided only at the global image-report level. This poses a central challenge: prior approaches spread weight densely across all patches rather than concentrating on the sparse subset relevant to a given query. To address this, we present GLINT (Gated Language-Image alignmeNT), a framework that models this sparse correspondence. On the alignment side, we introduce Sparsely Gated Alignment, a novel architecture in which a sigmoid gate over a separate gate embedding space activates only the patches relevant to each textual query, yielding sparsity. On the representation side, we add Dense Feature Regularization, which anchors the trainable encoder's intermediate features to a frozen self-supervised learning (SSL) teacher, preserving the fine-grained patch features that the gate relies on. The same recipe applies to both 2D chest X-ray (CXR) and 3D chest computed tomography (CT), built with DINOv3 and V-JEPA 2.1, respectively. GLINT enables zero-shot classification, grounding, and segmentation from free-text queries, and to our knowledge is the first to demonstrate zero-shot segmentation on 3D CT volumes without mask supervision. Notably, the most pronounced gains arise on zero-shot grounding and segmentation, where sparse, query-specific localization is required, consistent with our design intent. In downstream evaluation, GLINT outperforms both SSL encoders and medical VLMs on classification, report generation, and segmentation. Code is available at https://github.com/deepnoid-ai/GLINT.
This paper argues that dataset structure is important in image recognition tasks (among other tasks). Specifically, we focus on the nature and genesis of correlational structure in the actual datasets upon which DNNs are trained. We argue that DNNs are implementing a widespread methodology in condensed matter physics and materials science that focuses on mesoscale correlation structures that live between fundamental atomic/molecular scales and continuum scales. Specifically, we argue that DNNs that are successful in image classification must be discovering high order correlation functions. It is well-known that DNNs successfully generalize in apparent contravention of standard statistical learning theory. We consider the implications of our discussion for this puzzle.
Much of the literature on structured image recognition has disproportionately focused on the comparison of classification algorithms. Rather than investigating which classifier performs best, this paper instead asks: what should a classifier know before it ever makes a prediction? In structured vision problems such as gesture recognition, facial expression categorization, and medical image analysis, discriminative information lies less in individual pixels and more in spatial relationships between semantic parts. Raw pixel spaces are high-dimensional, sensitive to nuisance variation, and often obfuscate the geometric structures that make visual tasks interpretable. Landmark extraction provides one form of dimension reduction, but it does not by itself determine the information preserved. This paper studies the post-landmark feature map as the central object of analysis and proposes a systematic framework for constructing and interpreting landmark-derived representations as an, informed, feature-based "dimension reduction" step. Using static hand gesture recognition as a case study, we evaluate coordinate, distance, angle, and hybrid representations through perturbation and ablation experiments. The results show that visually variable data exposes substantial gaps between raw coordinate features and their geometrically invariant counterparts, while hybrid representations achieve the strongest overall performance by combining complementary geometric components. These findings frame feature construction as a fundamental modeling decision and ultimately suggests that the question of what representation should a classifier learn from is one worth asking. The code used for feature construction and evaluation is available at https://github.com/ShivMaureeCWRU/Feature_based_dimension_reduction
Conformal prediction (CP) is a powerful framework for quantifying uncertainty in machine learning models, offering reliable predictions with finite-sample coverage guarantees. When applied to classification, CP produces a prediction set of possible labels that is guaranteed to contain the true label with high probability, regardless of the underlying classifier. However, standard CP treats classes as flat and unstructured, ignoring domain knowledge such as semantic relationships or hierarchical structure among class labels. This paper presents hierarchical conformal classification (HCC), an extension of CP that incorporates class hierarchies into both the structure and semantics of prediction sets. We formulate HCC as a constrained optimization problem whose solutions yield prediction sets composed of nodes at different levels of the hierarchy, while maintaining coverage guarantees. To address the combinatorial nature of the problem, we formally show that a much smaller, well-structured subset of candidate solutions suffices to ensure coverage while upholding optimality. An empirical evaluation on three new benchmarks consisting of audio, image, and text data highlights the advantages of our approach, and a user study shows that annotators significantly prefer hierarchical over flat prediction sets.
Self-supervised learning (SSL) typically relies on a backbone encoder followed by a small multilayer perceptron (MLP) projection head, which is conventionally discarded after training, while backbone quality is assessed via costly linear probing on labeled data. We argue that this approach including discarding the projector is rather computationally wasteful. Instead, we propose replacing the MLP head with a radial basis function network (RBFN), whose interpretable center and shape parameters can be exploited to judge representation quality without labels or a separate classifier. To this end, we introduce Scale-Normalized Separation (SNS), a novel label-free quality metric derived solely from the kernel centers and shapes learned during training. Across five canonical SSL architectures (MoCo, SimCLR, BYOL, SwAV and SimSiam) and four image classification datasets, we show that RBFN projection heads are competitive drop-in replacements for standard MLP projectors. We recommend constructing them with three RBF layers activated by the Gaussian radial basis function. Moreover, SNS exhibits strong to very strong positive correlation with established logistic regression metrics, demonstrating that a trained RBFN projector can act as a reliable proxy for backbone representation quality. We additionally publish a novel PyTorch compatible image classification dataset based on Google's Open Images V7 to facilitate reproducible research into representation learning.
Traffic sign recognition (TSR) systems deployed on embedded vehicular hardware face deployment effects that image-based benchmark evaluation does not capture, including frame-to-frame prediction instability in streaming video and load-dependent throughput under sustained operation. We present Edge-TSR, a continuous edge inference system for TSR on the NVIDIA Jetson Orin Nano that combines sparse detection, multi-object tracking, region-level fine-grained classification, and a lightweight track-aware temporal stabilization mechanism adding only 0.07~ms per sampled frame. Evaluating on the source videos of the INTSD benchmark, we find that the state-of-the-art TSR models lose 20-30 percentage points of mAP@50 when moving from image-based to streaming evaluation. Edge-TSR achieves the best streaming classification performance among all evaluated baselines, and temporal stabilization improves accuracy by 10.58 percentage points over per-frame inference. We characterize latency, throughput, and thermal behavior across sampling intervals and four driving conditions, showing that throughput is load-proportional (20 ms per classified region) and that every-frame inference is limited by throughput rather than thermal throttling. A 55-minute, 26 km vehicular deployment demonstrates sustained online operation at 16.18 FPS within safe thermal limits without cloud offload. We also release the annotated 50,732-frame streaming evaluation dataset and full implementation.
Distributed denial-of-service (DDoS) attacks threaten the availability of Internet of Things (IoT) infrastructures, especially in resource-constrained deployments. Although transfer learning models have shown promising detection accuracy, their reliability, computational feasibility, and interpretability in operational environments remain insufficiently explored. This study presents an explainability-aware empirical evaluation of seven pre-trained convolutional neural network architectures for multi-class IoT DDoS detection using the CICDDoS2019 dataset and an image-based traffic representation. The analysis integrates classification performance metrics, complementary statistical indicators (MCC, Cohen's kappa, and Youden Index), training and inference cost assessment, and interpretability evaluation using Grad-CAM and SHAP. Results indicate that DenseNet and MobileNet-based architectures achieve strong detection performance while demonstrating superior reliability and compact, class-consistent attribution patterns. MobileNet achieves the highest reported classification accuracy and several of the strongest reliability statistics, whereas DenseNet169 exhibits the most favorable qualitative interpretability characteristics based on Grad-CAM and SHAP. MobileNetV3 offers a potential trade-off between inference latency, classification performance, and qualitative interpretability for resource-constrained deployment. The findings emphasize the importance of combining performance, reliability, and explainability criteria when selecting deep learning models for IoT DDoS detection.
Current scaling law formulations suggest that increasing the number of unique samples in a dataset consistently improves model performance, while repeated exposure to the same samples provides diminishing value. In this work, we challenge this assumption for image self-distillation models. We train iBOT models on random subsets of seven datasets spanning satellite imagery, driving scenes, egocentric video, and natural images, and evaluate their frozen representations on downstream tasks. We propose a density-interaction scaling law that combines unique-image count $U$ and repetition $R$, fitting it on runs with at most 50 million image presentations and testing it at larger budgets $C$. On Maxar dataset, a random subset containing approximately 1.5% of the training pool improves average nearest-neighbor classification accuracy by 3.11 percentage points over training on the full pool at the same budget of 200 million image presentations. Smaller subsets also outperform the full pool on SATLAS and Ego4D, whereas ImageNet favors larger pools at higher budgets. At a fixed subset size, further training can improve performance, lead to a plateau, or reduce performance. Our law captures smaller-subset advantages that the compared baselines cannot represent and improves prediction at higher budgets on Maxar, SATLAS, and Ego4D under nearest-neighbor evaluation. These findings show that both the number of unique images and how often they are repeated matter when allocating a training budget.
Vision Transformers (ViTs) have achieved state-of-the-art performance across a range of computer vision tasks, mainly thanks to the self-attention mechanism. However, its complexity, increasing quadratically with the number of tokens, remains the major obstacle to ViT efficiency and deployment at scale. Token merging reduces this cost by aggregating redundant tokens. Yet existing methods are typically evaluated within a single architecture, leaving open whether their effectiveness stems from the merging mechanism itself or from the specific decoder they are paired with. We extend Graph-Guided Token Merging (G2TM), a single module inserted early in a ViT-based network, beyond its original Segmenter setting. We evaluate G2TM across three semantic segmentation frameworks (Segmenter, SETR, EoMT) and three decoder families (Linear, Transformer-, convolution-based), as well as standard ViT image classification. Our results show that G2TM's behavior and accuracy-efficiency trade-off are consistent across every tested architecture for a given backbone size, indicating that its effectiveness is a property of the encoder rather than the decoder. G2TM also generalizes well to image classification, achieving an even smaller degradation in accuracy compared to semantic segmentation. We further find that G2TM's optimal hyperparameters, resulting in a consistent drop in GFLOPs of 22-47% and an increase in throughput by up to 74% for segmentation models on ADE20K dataset, depend primarily on the backbone's pre-training recipe and on the target dataset, rather than on the decoder choice.
Visual foundation models learn rich pathological morphology but lack explicit concept-level supervision. Vision--language pretraining can provide this supervision but requires costly image--text pairs. Here, we introduce GloSAGE, a semantic self-distillation framework that uses lesion profiles predicted by a pretrained visual teacher to construct structured pathology descriptions. Reliability-guided selection identifies image--text anchors, while Knowledge-Structured Pathology Text Synthesis and Evidence-Guided Negative-pair Construction account for partially observed lesions. Image--text alignment trains projection layers while retaining the frozen visual encoder; an intermediate Semantic Transfer Layer supplies features for downstream tasks. Developed on GloMillion and evaluated across seven glomerulus- and case-level cohorts, GloSAGE supported zero-shot lesion recognition, image--text retrieval, hard-negative discrimination and concept-intervention analyses. It outperformed its source visual model in most lesion-recognition tasks and improved lesion grading, while supporting competitive case-level disease classification, injury grading and baseline renal function estimation. Ablations supported the combined contributions of knowledge-structured text and evidence-guided negatives, and the intermediate transfer layer improved mean lesion-recognition ROC-AUC relative to the complete projection output. A frozen language model conditioned on GloSAGE features generated glomerular descriptions. These findings suggest that teacher-derived lesion profiles can provide annotation-efficient semantic supervision for extending a glomerular visual foundation model.