Deepfake detection models often struggle with generalization to unseen datasets, manifesting as misclassifying real instances as fake in target domains. This is primarily due to an overreliance on forgery artifacts and a limited understanding of real faces. To address this challenge, we propose a novel approach RealID to enhance generalization by learning a comprehensive concept of real faces while assessing the probabilities of belonging to the real and fake classes independently. RealID comprises two key modules: the Real Concept Capture Module (RealC^2) and the Independent Dual-Decision Classifier (IDC). With the assistance of a Multi-Real Memory, RealC^2 maintains various prototypes for real faces, allowing the model to capture a comprehensive concept of real class. Meanwhile, IDC redefines the classification strategy by making independent decisions based on the concept of the real class and the presence of forgery artifacts. Through the combined effect of the above modules, the influence of forgery-irrelevant patterns is alleviated, and extensive experiments on five widely used datasets demonstrate that RealID significantly outperforms existing state-of-the-art methods, achieving a 1.74% improvement in average accuracy.
论文检索
输入标题、作者或关键词，从 7,876 篇学术成果中精准定位
Cross-modal alignment is an effective approach to improving visual classification. Existing studies typically enforce a one-step mapping that uses deep neural networks to project the visual features to mimic the distribution of textual features. However, they typically face difficulties in finding such a projection due to the two modalities in both the distribution of class-wise samples and the range of their feature values. To address this issue, this paper proposes a novel Semantic-Space-Intervened Diffusive Alignment method, termed SeDA, models a semantic space as a bridge in the visual-to-textual projection, considering both types of features share the same class-level information in classification. More importantly, a bi-stage diffusion framework is developed to enable the progressive alignment between the two modalities. Specifically, SeDA first employs a Diffusion-Controlled Semantic Learner to model the semantic feature space of visual features by constraining the interactive features of the diffusion model and the category centers of visual features. In the later stage of SeDA, the Diffusion-Controlled Semantic Translator focuses on learning the distribution of textual features from the semantic space. Meanwhile, the Progressive Feature Interaction Network introduces stepwise feature interactions at each alignment step, progressively integrating textual information into mapped features. Experimental results show that SeDA achieves stronger cross-modal feature alignment, leading to superior performance over existing methods across multiple scenarios.
Traditional unsupervised domain adaptation (UDA) struggles to extract rich semantics due to backbone limitations. Recent large-scale pre-trained visual-language models (VLMs) have shown strong zero-shot learning capabilities in UDA tasks. However, directly using VLMs results in a mixture of semantic and domain-specific information, complicating knowledge transfer. Complex scenes with subtle semantic differences are prone to misclassification, which in turn can result in the loss of features that are crucial for distinguishing between classes. To address these challenges, we propose a novel counterfactual knowledge maintenance UDA framework. Specifically, we employ counterfactual disentanglement to separate the representation of semantic information from domain features, thereby reducing domain bias. Furthermore, to clarify ambiguous visual information specific to classes, we maintain the discriminative knowledge of both visual and textual information. This approach synergistically leverages multimodal information to preserve modality-specific distinguishable features. We conducted extensive experimental evaluations on several public datasets to demonstrate the effectiveness of our method. The source code is available at https://github.com/LiYaolab/CMKUDA
Compositional zero-shot learning (CZSL) is to recognize unseen attribute-object compositions by learning from seen compositions. The distribution shift between unseen compositions and seen compositions poses challenges to CZSL models, especially when test images are mixed with both seen and unseen compositions. The challenge will be addressed more easily if a model can distinguish unseen/seen compositions and treat them with specific recognition strategies. However, identifying images with unseen compositions is non-trivial, considering that unseen compositions are absent in training and usually contain only subtle differences from seen compositions. In this paper, we propose a novel compositional zero-shot learning method called COMO, which composes outliers in training for distinguishing seen and unseen compositions and further applying specific strategies for them. Specifically, we compose attribute-object representations for unseen compositions based on primitive representations of training images as outliers to enable the model to identify unseen compositions in inference. At test time, the method distinguishes images containing seen/unseen compositions and uses different weights for composition classification and primitive classification to recognize seen/unseen compositions. Experimental results on three datasets show the effectiveness of our method in both the closed-world setting and the open-world setting.
Cross-view geo-localization is a crucial task with diverse applications, yet it remains challenging due to the significant variations in viewpoints and visual appearances between images from different perspectives. While recent advancements have been made, existing methods often suffer from high model complexity, excessive resource consumption, and the impact of sample learning difficulty on optimization. To overcome these limitations, we optimize the Vision Mamba (Vim) model, built on a State Space Model (SSM) architecture, by replacing the traditional classification head with Channel Group Pooling (CGP) for efficient feature integration. This optimization reduces model parameters by 1.5% and computational complexity by 0.4%. Additionally, we propose a novel Dynamic Weighted Batch-tuple Loss (DWBL) to dynamically adjust the weighting of negative samples, improving model performance. By combining CGP and DWBL, we develop an efficient end-to-end network, VimGeo, which achieves state-of-the-art performance with enhanced computational efficiency. Specifically, VimGeo achieves a Recall@1 of 81.67% on the CVACT_test dataset, outperforming prior approaches. Extensive experiments on CVUSA, CVACT, and VIGOR datasets validate VimGeo's effectiveness and competitiveness in cross-view geo-localization tasks, achieving the leading results among sequence modeling-based methods. The implementation is available at: https://github.com/VimGeoTeam/VimGeo.
Adapting pre-trained vision-language models to downstream tasks has emerged as a novel paradigm for zero-shot learning. Existing test-time adaptation (TTA) methods such as TPT attempt to fine-tune visual or textual representations to accommodate downstream tasks but still require expensive optimization costs. To this end, Training-free Dynamic Adapter (TDA) maintains a cache containing visual features for each category in a parameter-free manner and measures sample confidence based on prediction entropy of test samples. Inspired by TDA, this work aims to develop the first training-free adapter for zero-shot video classification. Capturing the intrinsic temporal relationships within video data to construct and maintain the video cache is key to extending TDA to the video domain. In this work, we propose a reliable and diverse Hierarchical Adapter for zero-shot video classification, which consists of Frame-level Cache Refiner and Video-level Cache Updater. Before each video sample enters the corresponding cache, it needs to be refined at frame level based on prediction entropy and temporal probability difference. Due to the limited capacity of the cache, we update the cache during inference based on the principle of diversity. Experiments on four popular video classification benchmarks demonstrate the effectiveness of Hierarchical Adapter. The code is available at https://github.com/Gwxer/Hierarchical-Adapter.
Ordinal regression bridges regression and classification by assigning objects to ordered classes. While human experts rely on discriminative patch-level features for decisions, current approaches are limited by the availability of only image-level ordinal labels, overlooking fine-grained patch-level characteristics. In this paper, we propose a Dual-level Fuzzy Learning with Patch Guidance framework, named DFPG that learns precise feature-based grading boundaries from ambiguous ordinal labels, with patch-level supervision. Specifically, we propose patch-labeling and filtering strategies to enable the model to focus on patch-level features exclusively with only image-level ordinal labels available. We further design a dual-level fuzzy learning module, which leverages fuzzy logic to quantitatively capture and handle label ambiguity from both patch-wise and channel-wise perspectives. Extensive experiments on various image ordinal regression datasets demonstrate the superiority of our proposed method, further confirming its ability in distinguishing samples from difficult-to-classify categories. The code is available at https://github.com/ZJUMAI/DFPG-ord.
Enhancing Mixture of Experts with Independent and Collaborative Learning for Long-Tail Visual Recognition
PDF ↗Deep neural networks (DNNs) face substantial challenges in Long-Tail Visual Recognition (LTVR) due to the inherent class imbalances in real-world data distributions. The Mixture of Experts (MoE) framework has emerged as a promising approach to addressing these issues. However, in MoE systems, experts are typically trained to optimize a collective objective, often neglecting the individual optimality of each expert. This individual optimality usually contributes to the overall performance, as the goals of different experts are not mutually exclusive. We propose the Independent and Collaborative Learning (ICL) framework to optimize each expert independently while ensuring global optimality. First, Diverse Optimization Learning (DOL) is introduced to enhance expert diversity and individual performance. Then, we conceptualize experts as parallel circuit branches and introduce Competition and Collaboration Learning (CoL). Competition Learning amplifies the gradients of better-performing experts to preserve individual optimality, and Collaboration Learning encourages collaboration through mutual distillation to enhance optimal knowledge sharing. ICL achieves state-of-the-art accuracy in experiments on CIFAR-100/10-LT, ImageNet-LT, and iNaturalist 2018, respectively. Our code is available at https://github.com/PolarisLight/ICL.
A Medical Image Classification Network Based on Multi-View Consistent Momentum Contrastive Learning
PDF ↗Due to variations in imaging conditions, images often exhibit discrepancies in color reproduction. Furthermore, motion-induced blur can lead to edge degradation, making color sensitivity and edge blurriness two prevalent and challenging issues in both natural image processing and medical image analysis. To address these challenges, we propose a model termed the Three-View Consistency Mo-mentum Contrastive with Sobel Operator (SVCMC). Specifically, we first design a three-view momen-tum-update architecture that employs a So-bel-augmented ResNet as the backbone. We then introduce a novel contrastive loss, referred to as the Three-View Consistency Momentum Contrastive Loss. Next, to mitigate the oscillations and slow convergence commonly observed in contrastive learning, we construct a dynamic contrastive loss function that adapts in real time over the training process. Finally, we validated the superiority of our model on two medical image datasets and one natural image dataset, where its classification ac-curacy and convergence speed significantly out-performed existing state-of-the-art contrastive models.
Current feature fusion strategies often fail to adequately account for the influence of activation intensity across different scales on small object features, which impedes the effective detection of small objects. To address this limitation, we propose the Region-Adaptive Feature Disentanglement and Enhancement (RAFDE) strategy, which improves both downsampling and feature fusion by leveraging activation intensity variations at multiple scales. First, we introduce the Boundary Transitional Region-enhanced Downsampling (BTRD) module, which enhances boundary transitional regions containing both strongly and weakly activated features, thereby mitigating the loss of crucial boundary information for small objects. Second, we present the Regional-Adaptive Feature Fusion (RAFF) module, which adaptively disentangles and fuses co-activated and uni-activated regions from adjacent levels into the current level, effectively reducing the risk of small objects being overwhelmed. Extensive experiments on several public datasets demonstrate that the RAFDE strategy is highly effective and outperforms state-of-the-art methods. The code is available at https://github.com/b-yanchao/RAFDE.git.
Graph learning models have been empirically proven to be vulnerable to backdoor threats, wherein adversaries submit trigger-embedded inputs to manipulate the model predictions. Current graph backdoor defenses manifest several limitations: 1) dependence on model-related details, 2) necessitation of additional fine-tuning, and 3) reliance on extra explainability tools, all of which are infeasible under stringent privacy policies. To address those limitations, we propose GraphProt, a certified black-box defense method to suppress backdoor attacks on GNN-based graph classifiers. Our GraphProt operates in a model-agnostic manner and solely leverages graph input. Specifically, GraphProt first introduces designed topology-feature-filtration to mitigate graph anomalies. Subsequently, subgraphs are sampled via a formulated strategy integrating topology and features, followed by a robust model inference through a majority vote-based subgraph prediction ensemble. Our results across benchmark attacks and datasets show GraphProt effectively reduces attack success rates while preserving regular graph classification accuracy.
Real‐world datasets usually contain multiple attributes, making it essential to ensure fairness across all of them simultaneously. However, different attributes may vary in difficulty, and no existing approaches have effectively addressed this issue. Consequently, an attribute‐adaptive strategy is needed to achieve fairness for all attributes. Multi‐task Learning (MTL) leverages shared information to optimize multiple tasks concurrently, while Sparsely‐Gated Mixture‐of‐Experts (SMoE) can dynamically allocate computational resources to the most needed tasks. In this work, we formulate multi‐attribute fairness issue as an MTL problem and employ SMoE to achieve desirable performance across all attributes simultaneously. We first analyze the feasibility and find the potentiality by formalizing multi-attribute fairness problem into a MTL problem and mitigating it by using SMoE. However, vanilla SMoE could lead to over-utilization problem which causes sub-optimal performance. We then proposed an innovative SMoE framework for multi-attribute fair image classification, which further improves multi-attribute fairness by redesigning the MoE layer and routing policy with fairness consideration. Extensive experiments demonstrated the effectiveness. Taking a DeiT-Small as the backbone, we achieve 77.25% and 86.01% accuracy on the ISIC2019 and CelebA dataset respectively with Multi-attribute Predictive Quality Disparity (PQD) score of 0.801 and 0.787, beating current state-of-the-art methods Muffin, InfoFair and MultiFair.
MVP-CBM: Multi-layer Visual Preference-enhanced Concept Bottleneck Model for Explainable Medical Image Classification
PDF ↗The concept bottleneck model (CBM), as a technique improving interpretability via linking predictions to human-understandable concepts, makes high-risk and life-critical medical image classification credible. Typically, existing CBM methods associate the final layer of visual encoders with concepts to explain the model’s predictions. However, we empirically discover the phenomenon of concept preference variation, that is, the concepts are preferably associated with the features at different layers than those only at the final layer; yet a blind last-layer-based association neglects such a preference variation and thus weakens the accurate correspondences between features and concepts, impairing model interpretability. To address this issue, we propose a novel Multi-layer Visual Preference-enhanced Concept Bottleneck Model (MVP-CBM), which comprises two key novel modules: (1) intra-layer concept preference modeling, which captures the preferred association of different concepts with features at various visual layers, and (2) multi-layer concept sparse activation fusion, which sparsely aggregates concept activations from multiple layers to enhance performance. Thus, by explicitly modeling concept preferences, MVP-CBM can comprehensively leverage multi-layer visual information to provide a more nuanced and accurate explanation of model decisions. Extensive experiments on several public medical classification benchmarks demonstrate that MVP-CBM achieves state-of-the-art accuracy and interoperability, verifying its superiority. Code is available at https://github.com/wcj6/MVP-CBM.
Deepfake detection is increasingly crucial due to the rapid rise of AI-generated content. Existing methods achieve high performance relying on computationally intensive large models, making real-time detection on resource-constrained edge devices challenging. Given that deepfake detection is a binary classification task, there is potential for model compression and acceleration. In this paper, we propose a low-bit quantization framework for lightweight and efficient deepfake detection. The Connected Quantized Block extracts common forgery features via the quantized path and retains method-specific textures through the shortcut connections. Additionally, the Shifted Logarithmic Redistribution Quantizer mitigates information loss in near-zero domains by unfolding the unbalanced activations, enabling finer quantization granularity. Comprehensive experiments demonstrate that this new framework significantly reduces 10.8x computational costs and 12.4x storage requirements while maintaining high detection performance, even surpassing SOTA methods using less than 5% FLOPs, paving the way for efficient deepfake detection in resource-limited scenarios.
ASCENT-ViT: Attention-based Scale-aware Concept Learning Framework for Enhanced Alignment in Vision Transformers
PDF ↗As Vision Transformers (ViTs) are increasingly adopted in sensitive vision applications, there is a growing demand for improved interpretability. This has led to efforts to forward-align these models with carefully annotated abstract, human-understandable semantic entities - concepts. Concepts provide global rationales to the model predictions and can be quickly understood/intervened on by domain experts. Most current research focuses on designing model-agnostic, plug-and-play generic concept-based explainability modules that do not incorporate the inner workings of foundation models (e.g., inductive biases, scale invariance, etc.) during training. To alleviate this issue for ViTs, in this paper, we propose ASCENT-ViT, an attention-based, concept learning framework that effectively composes scale and position-aware representations from multiscale feature pyramids and ViT patch representations, respectively. Further, these representations are aligned with concept annotations through attention matrices - which incorporate spatial and global (semantic) concepts. ASCENT-ViT can be utilized as a classification head on top of standard ViT backbones for improved predictive performance and accurate and robust concept explanations as demonstrated on five datasets, including three widely used benchmarks (CUB, Pascal APY, Concept-MNIST) and two real-world datasets (AWA2, KITS). An appendix of the paper with more comprehensive results is available at https://arxiv.org/abs/2501.09221.
The challenge of delivering efficient explanations is a critical barrier that prevents the adoption of model explanations in real-world applications. Existing approaches often depend on extensive model queries for sample-level explanations or rely on expert's knowledge of specific model structures that trade general applicability for efficiency. To address these limitations, this paper introduces a novel framework Fast EXplanation (FEX) that represents attribution-based explanations via probability distributions, which are optimized by leveraging the policy gradient method. The proposed framework offers a robust, scalable solution for real-time, large-scale model explanations, bridging the gap between efficiency and applicability. We validate our framework on image and text classification tasks and the experiments demonstrate that our method reduces inference time by over 97 percent and memory usage by 70 percent compared to traditional model-agnostic approaches while maintaining high-quality explanations and broad applicability.
Established sampling protocols for 3D point cloud learning, such as Farthest Point Sampling (FPS) and Fixed Sample Size (FSS), have long been relied upon. However, real-world data often suffer from corruptions, such as sensor noise, which violates the benign data assumption in current protocols. As a result, these protocols are highly vulnerable to noise, posing significant safety risks in critical applications like autonomous driving. To address these issues, we propose an enhanced point cloud sampling protocol, PointSP, designed to improve robustness against point cloud corruptions. PointSP incorporates key point reweighting to mitigate outlier sensitivity and ensure the selection of representative points. It also introduces a local-global balanced downsampling strategy, which allows for scalable and adaptive sampling while maintaining geometric consistency. Additionally, a lightweight tangent plane interpolation method is used to preserve local geometry while enhancing the density of the point cloud. Unlike learning-based approaches that require additional model training, PointSP is architecture-agnostic, requiring no extra learning or modification to the network. This enables seamless integration into existing pipelines. Extensive experiments on synthetic and real-world corrupted datasets show that PointSP significantly improves the robustness and accuracy of point cloud classification, outperforming state-of-the-art methods across multiple benchmarks.
SuBiTO: Synopsis-based Training Optimization for Continuous Real-Time Neural Learning over Big Streaming Data
PDF ↗In machine learning applications over Big streaming Data, Neural Networks (NNs) are continuously and rapidly trained over voluminous data arriving at high speeds. As soon as a new version of the NN becomes available, it gets deployed for prediction purposes (e.g. classification). The real-time character of such applications greatly depends on the volume and velocity of the data streams, as well as the NN complexity. Training on large volume of ingested streams or using complex NNs, potentially increases accuracy, but may compromise the real-time character of those applications. In this work, we present SuBiTO, a framework that automatically and continuously learns the training time vs accuracy trade-offs as new data stream in and fine tunes: (i) the number, size and type of NN layers; (ii) the size of the ingested data via stream synopses specific parameters; and (iii) the number of training epochs. Finally, SuBiTO suggests optimal sets of such parameters and detects concept drifts, enabling the human operator adapt these parameters on-the-fly, at runtime.
Temporal Streaming Batch Principal Component Analysis for Time Series Classification (Student Abstract)
PDF ↗In multivariate time series classification, although current sequence analysis models have excellent classification capabilities, they show significant shortcomings when dealing with long sequence multivariate data. This paper focuses on optimizing model performance for long-sequence multivariate data by mitigating the impact of extended time series and multiple variables on the model. We propose a principal component analysis (PCA)-based temporal streaming compression and dimensionality reduction algorithm for time series data (temporal streaming batch PCA, TSBPCA), which continuously updates the compact representation of the entire sequence through streaming PCA time estimation with time block updates, enhancing the data representation capability of a range of sequence analysis models.We evaluated this method using various models on five datasets, and the experimental results show that our method demonstrates outstanding performance in both classification accuracy and time efficiency.
Domain-Informed Label Fusion Surpasses LLMs in Free-Living Activity Classification (Student Abstract)
PDF ↗FuSE-MET addresses critical challenges in deploying human activity recognition (HAR) systems in uncontrolled environments by effectively managing noisy labels, sparse data, and undefined activity vocabularies. By integrating BERT-based word embeddings with domain-specific knowledge (i.e., MET values), FuSE-MET optimizes label merging, reducing label complexity and improving classification accuracy. Our approach outperforms the state-of-the-art techniques, including ChatGPT-4, by balancing semantic meaning and physical intensity.