Hossein Kashiani

Niloufar Alipour Talemi, Hossein Kashiani, Fatemeh Afghah

Multimodal In-Context Learning (ICL) has emerged as a practical inference paradigm for Multimodal Large Language Models, where a small set of interleaved image-text In-Context Demonstrations (ICDs) conditions the model to solve new tasks. Despite its flexibility, multimodal ICL incurs high inference latency and suffers from instability due to sensitivity to demonstration formatting, ordering, and content. To address these limitations, we propose Hyper-ICL, a lightweight, training-based framework for demonstration-free multimodal ICL that reconstructs demonstration effects directly without requiring ICDs at inference time. Hyper-ICL learns a parameter-efficient low-rank logit-level adapter that calibrates attention distributions to better match demonstration-induced attention redistribution. To capture how demonstration influence varies across queries, we introduce a query-adaptive modulation mechanism that adaptively controls intervention strength at token level across layers and heads based on the current query. Finally, we propose a layer-wise hyperbolic anchor distillation loss that aligns intermediate student features to a demonstration-conditioned teacher via Lorentz geodesic distance. This loss encourages the student to reconstruct the demonstration-query relationships induced by ICDs. Extensive experiments across six different multimodal benchmarks (including VQAv2, OK-VQA, and COCO Caption) demonstrate that Hyper-ICL consistently improves accuracy and stability over vanilla ICL and existing state-of-the-art methods.

Omid Nejati Manzari, Hamid Ahmadabadi, Hossein Kashiani et al.

Convolutional Neural Networks (CNNs) have advanced existing medical systems for automatic disease diagnosis. However, there are still concerns about the reliability of deep medical diagnosis systems against the potential threats of adversarial attacks since inaccurate diagnosis could lead to disastrous consequences in the safety realm. In this study, we propose a highly robust yet efficient CNN-Transformer hybrid model which is equipped with the locality of CNNs as well as the global connectivity of vision Transformers. To mitigate the high quadratic complexity of the self-attention mechanism while jointly attending to information in various representation subspaces, we construct our attention mechanism by means of an efficient convolution operation. Moreover, to alleviate the fragility of our Transformer model against adversarial attacks, we attempt to learn smoother decision boundaries. To this end, we augment the shape information of an image in the high-level feature space by permuting the feature mean and variance within mini-batches. With less computational complexity, our proposed hybrid model demonstrates its high robustness and generalization ability compared to the state-of-the-art studies on a large-scale collection of standardized MedMNIST-2D datasets.

Atefeh Khoshkhahtinat, Ali Zafari, Piyush M. Mehta et al.

Transform and entropy models are the two core components in deep image compression neural networks. Most existing learning-based image compression methods utilize convolutional-based transform, which lacks the ability to model long-range dependencies, primarily due to the limited receptive field of the convolution operation. To address this limitation, we propose a Transformer-based nonlinear transform. This transform has the remarkable ability to efficiently capture both local and global information from the input image, leading to a more decorrelated latent representation. In addition, we introduce a novel entropy model that incorporates two different hyperpriors to model cross-channel and spatial dependencies of the latent representation. To further improve the entropy model, we add a global context that leverages distant relationships to predict the current latent more accurately. This global context employs a causal attention mechanism to extract long-range information in a content-dependent manner. Our experiments show that our proposed framework performs better than the state-of-the-art methods in terms of rate-distortion performance.

Hossein Kashiani, Shoaib Meraj Sami, Sobhan Soleymani et al.

Although a substantial amount of studies is dedicated to morph detection, most of them fail to generalize for morph faces outside of their training paradigm. Moreover, recent morph detection methods are highly vulnerable to adversarial attacks. In this paper, we intend to learn a morph detection model with high generalization to a wide range of morphing attacks and high robustness against different adversarial attacks. To this aim, we develop an ensemble of convolutional neural networks (CNNs) and Transformer models to benefit from their capabilities simultaneously. To improve the robust accuracy of the ensemble model, we employ multi-perturbation adversarial training and generate adversarial examples with high transferability for several single models. Our exhaustive evaluations demonstrate that the proposed robust ensemble model generalizes to several morphing attacks and face datasets. In addition, we validate that our robust ensemble model gain better robustness against several adversarial attacks while outperforming the state-of-the-art studies.

Niloufar Alipour Talemi, Hossein Kashiani, Sahar Rahimi Malakshan et al.

In this paper, we present a new multi-branch neural network that simultaneously performs soft biometric (SB) prediction as an auxiliary modality and face recognition (FR) as the main task. Our proposed network named AAFace utilizes SB attributes to enhance the discriminative ability of FR representation. To achieve this goal, we propose an attribute-aware attentional integration (AAI) module to perform weighted integration of FR with SB feature maps. Our proposed AAI module is not only fully context-aware but also capable of learning complex relationships between input features by means of the sequential multi-scale channel and spatial sub-modules. Experimental results verify the superiority of our proposed network compared with the state-of-the-art (SoTA) SB prediction and FR methods.

Mohammad Saeed Ebrahimi Saadabadi, Sahar Rahimi Malakshan, Hossein Kashiani et al.

In recent years, deep face recognition methods have demonstrated impressive results on in-the-wild datasets. However, these methods have shown a significant decline in performance when applied to real-world low-resolution benchmarks like TinyFace or SCFace. To address this challenge, we propose a novel classification consistency knowledge distillation approach that transfers the learned classifier from a high-resolution model to a low-resolution network. This approach helps in finding discriminative representations for low-resolution instances. To further improve the performance, we designed a knowledge distillation loss using the adaptive angular penalty inspired by the success of the popular angular margin loss function. The adaptive penalty reduces overfitting on low-resolution samples and alleviates the convergence issue of the model integrated with data augmentation. Additionally, we utilize an asymmetric cross-resolution learning approach based on the state-of-the-art semi-supervised representation learning paradigm to improve discriminability on low-resolution instances and prevent them from forming a cluster. Our proposed method outperforms state-of-the-art approaches on low-resolution benchmarks, with a three percent improvement on TinyFace while maintaining performance on high-resolution benchmarks.

Hossein Kashiani, Niloufar Alipour Talemi, Mohammad Saeed Ebrahimi Saadabadi et al.

Though recent studies have made significant progress in morph attack detection by virtue of deep neural networks, they often fail to generalize well to unseen morph attacks. With numerous morph attacks emerging frequently, generalizable morph attack detection has gained significant attention. This paper focuses on enhancing the generalization capability of morph attack detection from the perspective of consistency regularization. Consistency regularization operates under the premise that generalizable morph attack detection should output consistent predictions irrespective of the possible variations that may occur in the input space. In this work, to reach this objective, two simple yet effective morph-wise augmentations are proposed to explore a wide space of realistic morph transformations in our consistency regularization. Then, the model is regularized to learn consistently at the logit as well as embedding levels across a wide range of morph-wise augmented images. The proposed consistency regularization aligns the abstraction in the hidden layers of our model across the morph attack images which are generated from diverse domains in the wild. Experimental results demonstrate the superior generalization and robustness performance of our proposed method compared to the state-of-the-art studies.

Omid Nejati Manzari, Hossein Kashiani, Hojat Asgarian Dehkordi et al.

Vision transformers have been demonstrated to yield state-of-the-art results on a variety of computer vision tasks using attention-based networks. However, research works in transformers mostly do not investigate robustness/accuracy trade-off, and they still struggle to handle adversarial perturbations. In this paper, we explore the robustness of vision transformers against adversarial perturbations and try to enhance their robustness/accuracy trade-off in white box attack settings. To this end, we propose Locality iN Locality (LNL) transformer model. We prove that the locality introduction to LNL contributes to the robustness performance since it aggregates local information such as lines, edges, shapes, and even objects. In addition, to further improve the robustness performance, we encourage LNL to extract training signal from the moments (a.k.a., mean and standard deviation) and the normalized features. We validate the effectiveness and generality of LNL by achieving state-of-the-art results in terms of accuracy and robustness metrics on German Traffic Sign Recognition Benchmark (GTSRB) and Canadian Institute for Advanced Research (CIFAR-10). More specifically, for traffic sign classification, the proposed LNL yields gains of 1.1% and ~35% in terms of clean and robustness accuracy compared to the state-of-the-art studies.

Duncan McCain, Hossein Kashiani, Fatemeh Afghah

Visual anomaly detection in multi-class settings poses significant challenges due to the diversity of object categories, the scarcity of anomalous examples, and the presence of camouflaged defects. In this paper, we propose PromptMAD, a cross-modal prompting framework for unsupervised visual anomaly detection and localization that integrates semantic guidance through vision-language alignment. By leveraging CLIP-encoded text prompts describing both normal and anomalous class-specific characteristics, our method enriches visual reconstruction with semantic context, improving the detection of subtle and textural anomalies. To further address the challenge of class imbalance at the pixel level, we incorporate Focal loss function, which emphasizes hard-to-detect anomalous regions during training. Our architecture also includes a supervised segmentor that fuses multi-scale convolutional features with Transformer-based spatial attention and diffusion iterative refinement, yielding precise and high-resolution anomaly maps. Extensive experiments on the MVTec-AD dataset demonstrate that our method achieves state-of-the-art pixel-level performance, improving mean AUC to 98.35% and AP to 66.54%, while maintaining efficiency across diverse categories.

Niloufar Alipour Talemi, Hossein Kashiani, Nasser M. Nasrabadi

Although face recognition (FR) has achieved great success in recent years, it is still challenging to accurately recognize faces in low-quality images due to the obscured facial details. Nevertheless, it is often feasible to make predictions about specific soft biometric (SB) attributes, such as gender, and baldness even in dealing with low-quality images. In this paper, we propose a novel multi-branch neural network that leverages SB attribute information to boost the performance of FR. To this end, we propose a cross-attribute-guided transformer fusion (CATF) module that effectively captures the long-range dependencies and relationships between FR and SB feature representations. The synergy created by the reciprocal flow of information in the dual cross-attention operations of the proposed CATF module enhances the performance of FR. Furthermore, we introduce a novel self-attention distillation framework that effectively highlights crucial facial regions, such as landmarks by aligning low-quality images with those of their high-quality counterparts in the feature space. The proposed self-attention distillation regularizes our network to learn a unified quality-invariant feature representation in unconstrained environments. We conduct extensive experiments on various FR benchmarks varying in quality. Experimental results demonstrate the superiority of our FR method compared to state-of-the-art FR studies.

Hossein Kashiani, Niloufar Alipour Talemi, Fatemeh Afghah

Recent advancements in anomaly detection have shifted focus towards Multi-class Unified Anomaly Detection (MUAD), offering more scalable and practical alternatives compared to traditional one-class-one-model approaches. However, existing MUAD methods often suffer from inter-class interference and are highly susceptible to domain shifts, leading to substantial performance degradation in real-world applications. In this paper, we propose a novel robust prompt-driven MUAD framework, called ROADS, to address these challenges. ROADS employs a hierarchical class-aware prompt integration mechanism that dynamically encodes class-specific information into our anomaly detector to mitigate interference among anomaly classes. Additionally, ROADS incorporates a domain adapter to enhance robustness against domain shifts by learning domain-invariant representations. Extensive experiments on MVTec-AD and VISA datasets demonstrate that ROADS surpasses state-of-the-art methods in both anomaly detection and localization, with notable improvements in out-of-distribution settings.

Hossein Kashiani, Niloufar Alipour Talemi, Fatemeh Afghah

Deepfake detectors often struggle to generalize to novel forgery types due to biases learned from limited training data. In this paper, we identify a new type of model bias in the frequency domain, termed spectral bias, where detectors overly rely on specific frequency bands, restricting their ability to generalize across unseen forgeries. To address this, we propose FreqDebias, a frequency debiasing framework that mitigates spectral bias through two complementary strategies. First, we introduce a novel Forgery Mixup (Fo-Mixup) augmentation, which dynamically diversifies frequency characteristics of training samples. Second, we incorporate a dual consistency regularization (CR), which enforces both local consistency using class activation maps (CAMs) and global consistency through a von Mises-Fisher (vMF) distribution on a hyperspherical embedding space. This dual CR mitigates over-reliance on certain frequency components by promoting consistent representation learning under both local and global supervision. Extensive experiments show that FreqDebias significantly enhances cross-domain generalization and outperforms state-of-the-art methods in both cross-domain and in-domain settings.

Niloufar Alipour Talemi, Hossein Kashiani, Fatemeh Afghah

Pre-trained Vision-language (VL) models, such as CLIP, have shown significant generalization ability to downstream tasks, even with minimal fine-tuning. While prompt learning has emerged as an effective strategy to adapt pre-trained VL models for downstream tasks, current approaches frequently encounter severe overfitting to specific downstream data distributions. This overfitting constrains the original behavior of the VL models to generalize to new domains or unseen classes, posing a critical challenge in enhancing the adaptability and generalization of VL models. To address this limitation, we propose Style-Pro, a novel style-guided prompt learning framework that mitigates overfitting and preserves the zero-shot generalization capabilities of CLIP. Style-Pro employs learnable style bases to synthesize diverse distribution shifts, guided by two specialized loss functions that ensure style diversity and content integrity. Then, to minimize discrepancies between unseen domains and the source domain, Style-Pro maps the unseen styles into the known style representation space as a weighted combination of style bases. Moreover, to maintain consistency between the style-shifted prompted model and the original frozen CLIP, Style-Pro introduces consistency constraints to preserve alignment in the learned embeddings, minimizing deviation during adaptation to downstream tasks. Extensive experiments across 11 benchmark datasets demonstrate the effectiveness of Style-Pro, consistently surpassing state-of-the-art methods in various settings, including base-to-new generalization, cross-dataset transfer, and domain generalization.

Niloufar Alipour Talemi, Hossein Kashiani, Hossein R. Nowdeh et al.

Prompt learning has emerged as a powerful paradigm for adapting vision-language models such as CLIP to downstream tasks. However, existing methods often overfit to seen data, leading to significant performance degradation when generalizing to novel classes or unseen domains. To address this limitation, we propose DiSa, a Directional Saliency-Aware Prompt Learning framework that integrates two complementary regularization strategies to enhance generalization. First, our Cross-Interactive Regularization (CIR) fosters cross-modal alignment by enabling cooperative learning between prompted and frozen encoders. Within CIR, a saliency-aware masking strategy guides the image encoder to prioritize semantically critical image regions, reducing reliance on less informative patches. Second, we introduce a directional regularization strategy that aligns visual embeddings with class-wise prototype features in a directional manner to prioritize consistency in feature orientation over strict proximity. This approach ensures robust generalization by leveraging stable prototype directions derived from class-mean statistics. Extensive evaluations on 11 diverse image classification benchmarks demonstrate that DiSa consistently outperforms state-of-the-art prompt learning methods across various settings, including base-to-novel generalization, cross-dataset transfer, domain generalization, and few-shot learning.

Hojat Asgarian Dehkordi, Ali Soltani Nezhad, Hossein Kashiani et al.

In still image human action recognition, existing studies have mainly leveraged extra bounding box information along with class labels to mitigate the lack of temporal information in still images; however, preparing extra data with manual annotation is time-consuming and also prone to human errors. Moreover, the existing studies have not addressed action recognition with long-tailed distribution. In this paper, we propose a two-phase multi-expert classification method for human action recognition to cope with long-tailed distribution by means of super-class learning and without any extra information. To choose the best configuration for each super-class and characterize inter-class dependency between different action classes, we propose a novel Graph-Based Class Selection (GCS) algorithm. In the proposed approach, a coarse-grained phase selects the most relevant fine-grained experts. Then, the fine-grained experts encode the intricate details within each super-class so that the inter-class variation increases. Extensive experimental evaluations are conducted on various public human action recognition datasets, including Stanford40, Pascal VOC 2012 Action, BU101+, and IHAR datasets. The experimental results demonstrate that the proposed method yields promising improvements. To be more specific, in IHAR, Sanford40, Pascal VOC 2012 Action, and BU101+ benchmarks, the proposed approach outperforms the state-of-the-art studies by 8.92%, 0.41%, 0.66%, and 2.11 % with much less computational cost and without any auxiliary annotation information. Besides, it is proven that in addressing action recognition with long-tailed distribution, the proposed method outperforms its counterparts by a significant margin.

Hossein Kashiani, Amir Abbas Hamidi Imani, Shahriar Baradaran Shokouhi et al.

Online learning policy makes visual trackers more robust against different distortions through learning domain-specific cues. However, the trackers adopting this policy fail to fully leverage the discriminative context of the background areas. Moreover, owing to the lack of sufficient data at each time step, the online learning approach can also make the trackers prone to over-fitting to the background regions. In this paper, we propose a domain adaptation approach to strengthen the contributions of the semantic background context. The domain adaptation approach is backboned with only an off-the-shelf deep model. The strength of the proposed approach comes from its discriminative ability to handle severe occlusion and background clutter challenges. We further introduce a cost-sensitive loss alleviating the dominance of non-semantic background candidates over the semantic candidates, thereby dealing with the data imbalance issue. Experimental results demonstrate that our tracker achieves competitive results at real-time speed compared to the state-of-the-art trackers.

Hossein Kashiani, Shahriar B. Shokouhi

Recently, convolutional neural network (CNN) has attracted much attention in different areas of computer vision, due to its powerful abstract feature representation. Visual object tracking is one of the interesting and important areas in computer vision that achieves remarkable improvements in recent years. In this work, we aim to improve both the motion and observation models in visual object tracking by leveraging representation power of CNNs. To this end, a motion estimation network (named MEN) is utilized to seek the most likely locations of the target and prepare a further clue in addition to the previous target position. Hence the motion estimation would be enhanced by generating a small number of candidates near two plausible positions. The generated candidates are then fed into a trained Siamese network to detect the most probable candidate. Each candidate is compared to an adaptable buffer, which is updated under a predefined condition. To take into account the target appearance changes, a weighting CNN (called WCNN) adaptively assigns weights to the final similarity scores of the Siamese network using sequence-specific information. Evaluation results on well-known benchmark datasets (OTB100, OTB50 and OTB2013) prove that the proposed tracker outperforms the state-of-the-art competitors.