Ognjen Arandjelović

Yifei Qian, Xiaopeng Hong, Zhongliang Guo et al.

To alleviate the heavy annotation burden for training a reliable crowd counting model and thus make the model more practicable and accurate by being able to benefit from more data, this paper presents a new semi-supervised method based on the mean teacher framework. When there is a scarcity of labeled data available, the model is prone to overfit local patches. Within such contexts, the conventional approach of solely improving the accuracy of local patch predictions through unlabeled data proves inadequate. Consequently, we propose a more nuanced approach: fostering the model's intrinsic 'subitizing' capability. This ability allows the model to accurately estimate the count in regions by leveraging its understanding of the crowd scenes, mirroring the human cognitive process. To achieve this goal, we apply masking on unlabeled data, guiding the model to make predictions for these masked patches based on the holistic cues. Furthermore, to help with feature learning, herein we incorporate a fine-grained density classification task. Our method is general and applicable to most existing crowd counting methods as it doesn't have strict structural or loss constraints. In addition, we observe that the model trained with our framework exhibits a 'subitizing'-like behavior. It accurately predicts low-density regions with only a 'glance', while incorporating local details to predict high-density regions. Our method achieves the state-of-the-art performance, surpassing previous approaches by a large margin on challenging benchmarks such as ShanghaiTech A and UCF-QNRF. The code is available at: https://github.com/cha15yq/MRC-Crowd.

Georg Wölflein, Dyke Ferber, Asier R. Meneghetti et al.

Weakly supervised whole slide image classification is a key task in computational pathology, which involves predicting a slide-level label from a set of image patches constituting the slide. Constructing models to solve this task involves multiple design choices, often made without robust empirical or conclusive theoretical justification. To address this, we conduct a comprehensive benchmarking of feature extractors to answer three critical questions: 1) Is stain normalisation still a necessary preprocessing step? 2) Which feature extractors are best for downstream slide-level classification? 3) How does magnification affect downstream performance? Our study constitutes the most comprehensive evaluation of publicly available pathology feature extractors to date, involving more than 10,000 training runs across 14 feature extractors, 9 tasks, 5 datasets, 3 downstream architectures, 2 levels of magnification, and various preprocessing setups. Our findings challenge existing assumptions: 1) We observe empirically, and by analysing the latent space, that skipping stain normalisation and image augmentations does not degrade performance, while significantly reducing memory and computational demands. 2) We develop a novel evaluation metric to compare relative downstream performance, and show that the choice of feature extractor is the most consequential factor for downstream performance. 3) We find that lower-magnification slides are sufficient for accurate slide-level classification. Contrary to previous patch-level benchmarking studies, our approach emphasises clinical relevance by focusing on slide-level biomarker prediction tasks in a weakly supervised setting with external validation cohorts. Our findings stand to streamline digital pathology workflows by minimising preprocessing needs and informing the selection of feature extractors.

Zhongliang Guo, Chun Tong Lei, Lei Fang et al.

Recent advancements in Latent Diffusion Models (LDMs) have revolutionized image synthesis and manipulation, raising significant concerns about data misappropriation and intellectual property infringement. While adversarial attacks have been extensively explored as a protective measure against such misuse of generative AI, current approaches are severely limited by their heavy reliance on model-specific knowledge and substantial computational costs. Drawing inspiration from the posterior collapse phenomenon observed in VAE training, we propose the Posterior Collapse Attack (PCA), a novel framework for protecting images from unauthorized manipulation. Through comprehensive theoretical analysis and empirical validation, we identify two distinct collapse phenomena during VAE inference: diffusion collapse and concentration collapse. Based on this discovery, we design a unified loss function that can flexibly achieve both types of collapse through parameter adjustment, each corresponding to different protection objectives in preventing image manipulation. Our method significantly reduces dependence on model-specific knowledge by requiring access to only the VAE encoder, which constitutes less than 4\% of LDM parameters. Notably, PCA achieves prompt-invariant protection by operating on the VAE encoder before text conditioning occurs, eliminating the need for empty prompt optimization required by existing methods. This minimal requirement enables PCA to maintain adequate transferability across various VAE-based LDM architectures while effectively preventing unauthorized image editing. Extensive experiments show PCA outperforms existing techniques in protection effectiveness, computational efficiency (runtime and VRAM), and generalization across VAE-based LDM variants. Our code is available at https://github.com/ZhongliangGuo/PosteriorCollapseAttack.

Haseeb Nazki, Ognjen Arandjelović, InHwa Um et al.

Histopathology relies on the analysis of microscopic tissue images to diagnose disease. A crucial part of tissue preparation is staining whereby a dye is used to make the salient tissue components more distinguishable. However, differences in laboratory protocols and scanning devices result in significant confounding appearance variation in the corresponding images. This variation increases both human error and the inter-rater variability, as well as hinders the performance of automatic or semi-automatic methods. In the present paper we introduce an unsupervised adversarial network to translate (and hence normalize) whole slide images across multiple data acquisition domains. Our key contributions are: (i) an adversarial architecture which learns across multiple domains with a single generator-discriminator network using an information flow branch which optimizes for perceptual loss, and (ii) the inclusion of an additional feature extraction network during training which guides the transformation network to keep all the structural features in the tissue image intact. We: (i) demonstrate the effectiveness of the proposed method firstly on H\&E slides of 120 cases of kidney cancer, as well as (ii) show the benefits of the approach on more general problems, such as flexible illumination based natural image enhancement and light source adaptation.

Zhongliang Guo, Weiye Li, Yifei Qian et al.

In this paper, we tackle the challenge of white-box false positive adversarial attacks on contrastive loss based offline handwritten signature verification models. We propose a novel attack method that treats the attack as a style transfer between closely related but distinct writing styles. To guide the generation of deceptive images, we introduce two new loss functions that enhance the attack success rate by perturbing the Euclidean distance between the embedding vectors of the original and synthesized samples, while ensuring minimal perturbations by reducing the difference between the generated image and the original image. Our method demonstrates state-of-the-art performance in white-box attacks on contrastive loss based offline handwritten signature verification models, as evidenced by our experiments. The key contributions of this paper include a novel false positive attack method, two new loss functions, effective style transfer in handwriting styles, and superior performance in white-box false positive attacks compared to other white-box attack methods.

Georg Wölflein, In Hwa Um, David J Harrison et al.

The presence and density of specific types of immune cells are important to understand a patient's immune response to cancer. However, immunofluorescence staining required to identify T cell subtypes is expensive, time-consuming, and rarely performed in clinical settings. We present a framework to virtually stain Hoechst images (which are cheap and widespread) with both CD3 and CD8 to identify T cell subtypes in clear cell renal cell carcinoma using generative adversarial networks. Our proposed method jointly learns both staining tasks, incentivising the network to incorporate mutually beneficial information from each task. We devise a novel metric to quantify the virtual staining quality, and use it to evaluate our method.

Liangfei Zhang, Yifei Qian, Ognjen Arandjelovic et al.

This paper discusses the benefits of incorporating multimodal data for improving latent emotion recognition accuracy, focusing on micro-expression (ME) and physiological signals (PS). The proposed approach presents a novel multimodal learning framework that combines ME and PS, including a 1D separable and mixable depthwise inception network, a standardised normal distribution weighted feature fusion method, and depth/physiology guided attention modules for multimodal learning. Experimental results show that the proposed approach outperforms the benchmark method, with the weighted fusion method and guided attention modules both contributing to enhanced performance.

Georg Wölflein, Dyke Ferber, Daniel Truhn et al.

Tool use has turned large language models (LLMs) into powerful agents that can perform complex multi-step tasks by dynamically utilising external software components. However, these tools must be implemented in advance by human developers, hindering the applicability of LLM agents in domains demanding large numbers of highly specialised tools, like in life sciences and medicine. Motivated by the growing trend of scientific studies accompanied by public code repositories, we propose ToolMaker, an agentic framework that autonomously transforms papers with code into LLM-compatible tools. Given a GitHub URL and short task description, ToolMaker autonomously installs dependencies and generates code to perform the task, using a closed-loop self-correction mechanism for debugging. To evaluate our approach, we introduce a benchmark comprising 15 complex computational tasks spanning various domains with over 100 unit tests to assess correctness and robustness. Our method correctly implements 80% of the tasks, substantially outperforming current state-of-the-art software engineering agents. ToolMaker therefore is a step towards fully autonomous agent-based scientific workflows. Our code and benchmark are publicly available at https://github.com/KatherLab/ToolMaker.

Georg Wölflein, Lucie Charlotte Magister, Pietro Liò et al.

Traditional supervised learning tasks require a label for every instance in the training set, but in many real-world applications, labels are only available for collections (bags) of instances. This problem setting, known as multiple instance learning (MIL), is particularly relevant in the medical domain, where high-resolution images are split into smaller patches, but labels apply to the image as a whole. Recent MIL models are able to capture correspondences between patches by employing self-attention, allowing them to weigh each patch differently based on all other patches in the bag. However, these approaches still do not consider the relative spatial relationships between patches within the larger image, which is especially important in computational pathology. To this end, we introduce a novel MIL model with distance-aware self-attention (DAS-MIL), which explicitly takes into account relative spatial information when modelling the interactions between patches. Unlike existing relative position representations for self-attention which are discrete, our approach introduces continuous distance-dependent terms into the computation of the attention weights, and is the first to apply relative position representations in the context of MIL. We evaluate our model on a custom MNIST-based MIL dataset that requires the consideration of relative spatial information, as well as on CAMELYON16, a publicly available cancer metastasis detection dataset, where we achieve a test AUROC score of 0.91. On both datasets, our model outperforms existing MIL approaches that employ absolute positional encodings, as well as existing relative position representation schemes applied to MIL. Our code is available at https://anonymous.4open.science/r/das-mil.

Zhongliang Guo, Yifei Qian, Yanli Li et al.

Adversarial attacks against computer vision systems have emerged as a critical research area that challenges the fundamental assumptions about neural network robustness and security. This comprehensive survey examines the evolving landscape of adversarial techniques, revealing their dual nature as both sophisticated security threats and valuable defensive tools. We provide a systematic analysis of adversarial attack methodologies across three primary domains: pixel-space attacks, physically realizable attacks, and latent-space attacks. Our investigation traces the technical evolution from early gradient-based methods such as FGSM and PGD to sophisticated optimization techniques incorporating momentum, adaptive step sizes, and advanced transferability mechanisms. We examine how physically realizable attacks have successfully bridged the gap between digital vulnerabilities and real-world threats through adversarial patches, 3D textures, and dynamic optical perturbations. Additionally, we explore the emergence of latent-space attacks that leverage semantic structure in internal representations to create more transferable and meaningful adversarial examples. Beyond traditional offensive applications, we investigate the constructive use of adversarial techniques for vulnerability assessment in biometric authentication systems and protection against malicious generative models. Our analysis reveals critical research gaps, particularly in neural style transfer protection and computational efficiency requirements. This survey contributes a comprehensive taxonomy, evolution analysis, and identification of future research directions, aiming to advance understanding of adversarial vulnerabilities and inform the development of more robust and trustworthy computer vision systems.

Andreas Hadjipieris, Neofytos Dimitriou, Ognjen Arandjelović

Port berthing sites are regions of high interest for monitoring and optimizing port operations. Data sourced from the Automatic Identification System (AIS) can be superimposed on berths enabling their real-time monitoring and revealing long-term utilization patterns. Ultimately, insights from multiple berths can uncover bottlenecks, and lead to the optimization of the underlying supply chain of the port and beyond. However, publicly available documentation of port berths, even when available, is frequently incomplete - e.g. there may be missing berths or inaccuracies such as incorrect boundary boxes - necessitating a more robust, data-driven approach to port berth localization. In this context, we propose an unsupervised spatial modeling method that leverages AIS data clustering and hyperparameter optimization to identify berthing sites. Trained on one month of freely available AIS data and evaluated across ports of varying sizes, our models significantly outperform competing methods, achieving a mean Bhattacharyya distance of 0.85 when comparing Gaussian Mixture Models (GMMs) trained on separate data splits, compared to 13.56 for the best existing method. Qualitative comparison with satellite images and existing berth labels further supports the superiority of our method, revealing more precise berth boundaries and improved spatial resolution across diverse port environments.

Zhongliang Guo, Junhao Dong, Yifei Qian et al.

Neural style transfer (NST) generates new images by combining the style of one image with the content of another. However, unauthorized NST can exploit artwork, raising concerns about artists' rights and motivating the development of proactive protection methods. We propose Locally Adaptive Adversarial Color Attack (LAACA), empowering artists to protect their artwork from unauthorized style transfer by processing before public release. By delving into the intricacies of human visual perception and the role of different frequency components, our method strategically introduces frequency-adaptive perturbations in the image. These perturbations significantly degrade the generation quality of NST while maintaining an acceptable level of visual change in the original image, ensuring that potential infringers are discouraged from using the protected artworks, because of its bad NST generation quality. Additionally, existing metrics often overlook the importance of color fidelity in evaluating color-mattered tasks, such as the quality of NST-generated images, which is crucial in the context of artistic works. To comprehensively assess the color-mattered tasks, we propose the Adversarial Color Distance Metric (ACDM), designed to quantify the color difference of images pre- and post-manipulations. Experimental results confirm that attacking NST using LAACA results in visually inferior style transfer, and the ACDM can efficiently measure color-mattered tasks. By providing artists with a tool to safeguard their intellectual property, our work relieves the socio-technical challenges posed by the misuse of NST in the art community.

Avelina Asada Hadji-Kyriacou, Ognjen Arandjelovic

This paper introduces a novel Parameter-Efficient Fine-Tuning (PEFT) framework for multi-modal, multi-task transfer learning with pre-trained language models. PEFT techniques such as LoRA, BitFit and IA3 have demonstrated comparable performance to full fine-tuning of pre-trained models for specific downstream tasks, all while demanding significantly fewer trainable parameters and reduced GPU memory consumption. However, in the context of multi-modal fine-tuning, the need for architectural modifications or full fine-tuning often becomes apparent. To address this we propose Context-PEFT, which learns different groups of adaptor parameters based on the token's domain. This approach enables LoRA-like weight injection without requiring additional architectural changes. Our method is evaluated on the COCO captioning task, where it outperforms full fine-tuning under similar data constraints while simultaneously offering a substantially more parameter-efficient and computationally economical solution.

Neofytos Dimitriou, Ognjen Arandjelovic

The shift towards end-to-end deep learning has brought unprecedented advances in many areas of computer vision. However, deep neural networks are trained on images with resolutions that rarely exceed $1,000 \times 1,000$ pixels. The growing use of scanners that create images with extremely high resolutions (average can be $100,000 \times 100,000$ pixels) thereby presents novel challenges to the field. Most of the published methods preprocess high-resolution images into a set of smaller patches, imposing an a priori belief on the best properties of the extracted patches (magnification, field of view, location, etc.). Herein, we introduce Magnifying Networks (MagNets) as an alternative deep learning solution for gigapixel image analysis that does not rely on a preprocessing stage nor requires the processing of billions of pixels. MagNets can learn to dynamically retrieve any part of a gigapixel image, at any magnification level and field of view, in an end-to-end fashion with minimal ground truth (a single global, slide-level label). Our results on the publicly available Camelyon16 and Camelyon17 datasets corroborate to the effectiveness and efficiency of MagNets and the proposed optimization framework for whole slide image classification. Importantly, MagNets process far less patches from each slide than any of the existing approaches ($10$ to $300$ times less).

Liangfei Zhang, Xiaopeng Hong, Ognjen Arandjelovic et al.

Being spontaneous, micro-expressions are useful in the inference of a person's true emotions even if an attempt is made to conceal them. Due to their short duration and low intensity, the recognition of micro-expressions is a difficult task in affective computing. The early work based on handcrafted spatio-temporal features which showed some promise, has recently been superseded by different deep learning approaches which now compete for the state of the art performance. Nevertheless, the problem of capturing both local and global spatio-temporal patterns remains challenging. To this end, herein we propose a novel spatio-temporal transformer architecture -- to the best of our knowledge, the first purely transformer based approach (i.e. void of any convolutional network use) for micro-expression recognition. The architecture comprises a spatial encoder which learns spatial patterns, a temporal aggregator for temporal dimension analysis, and a classification head. A comprehensive evaluation on three widely used spontaneous micro-expression data sets, namely SMIC-HS, CASME II and SAMM, shows that the proposed approach consistently outperforms the state of the art, and is the first framework in the published literature on micro-expression recognition to achieve the unweighted F1-score greater than 0.9 on any of the aforementioned data sets.

Jessica Cooper, In Hwa Um, Ognjen Arandjelović et al.

Multiplex immunofluorescence and immunohistochemistry benefit patients by allowing cancer pathologists to identify several proteins expressed on the surface of cells, enabling cell classification, better understanding of the tumour micro-environment, more accurate diagnoses, prognoses, and tailored immunotherapy based on the immune status of individual patients. However, they are expensive and time consuming processes which require complex staining and imaging techniques by expert technicians. Hoechst staining is much cheaper and easier to perform, but is not typically used in this case as it binds to DNA rather than to the proteins targeted by immunofluorescent techniques, and it was not previously thought possible to differentiate cells expressing these proteins based only on DNA morphology. In this work we show otherwise, training a deep convolutional neural network to identify cells expressing three proteins (T lymphocyte markers CD3 and CD8, and the B lymphocyte marker CD20) with greater than 90% precision and recall, from Hoechst 33342 stained tissue only. Our model learns previously unknown morphological features associated with expression of these proteins which can be used to accurately differentiate lymphocyte subtypes for use in key prognostic metrics such as assessment of immune cell infiltration,and thereby predict and improve patient outcomes without the need for costly multiplex immunofluorescence.

Jessica Cooper, Ognjen Arandjelović, David J Harrison

Understanding the predictions made by Artificial Intelligence (AI) systems is becoming more and more important as deep learning models are used for increasingly complex and high-stakes tasks. Saliency mapping -- a popular visual attribution method -- is one important tool for this, but existing formulations are limited by either computational cost or architectural constraints. We therefore propose Hierarchical Perturbation, a very fast and completely model-agnostic method for interpreting model predictions with robust saliency maps. Using standard benchmarks and datasets, we show that our saliency maps are of competitive or superior quality to those generated by existing model-agnostic methods -- and are over 20 times faster to compute.

Neofytos Dimitriou, Ognjen Arandjelovic

Batch normalization (BatchNorm) is an effective yet poorly understood technique for neural network optimization. It is often assumed that the degradation in BatchNorm performance to smaller batch sizes stems from it having to estimate layer statistics using smaller sample sizes. However, recently, Ghost normalization (GhostNorm), a variant of BatchNorm that explicitly uses smaller sample sizes for normalization, has been shown to improve upon BatchNorm in some datasets. Our contributions are: (i) we uncover a source of regularization that is unique to GhostNorm, and not simply an extension from BatchNorm, (ii) three types of GhostNorm implementations are described, two of which employ BatchNorm as the underlying normalization technique, (iii) by visualising the loss landscape of GhostNorm, we observe that GhostNorm consistently decreases the smoothness when compared to BatchNorm, (iv) we introduce Sequential Normalization (SeqNorm), and report superior performance over state-of-the-art methodologies on both CIFAR--10 and CIFAR--100 datasets.

Neofytos Dimitriou, Ognjen Arandjelović, Peter D Caie

The widespread adoption of whole slide imaging has increased the demand for effective and efficient gigapixel image analysis. Deep learning is at the forefront of computer vision, showcasing significant improvements over previous methodologies on visual understanding. However, whole slide images have billions of pixels and suffer from high morphological heterogeneity as well as from different types of artefacts. Collectively, these impede the conventional use of deep learning. For the clinical translation of deep learning solutions to become a reality, these challenges need to be addressed. In this paper, we review work on the interdisciplinary attempt of training deep neural networks using whole slide images, and highlight the different ideas underlying these methodologies.

Jessica Cooper, Ognjen Arandjelovic

In recent years, a range of problems within the broad umbrella of automatic, computer vision based analysis of ancient coins has been attracting an increasing amount of attention. Notwithstanding this research effort, the results achieved by the state of the art in the published literature remain poor and far from sufficiently well performing for any practical purpose. In the present paper we present a series of contributions which we believe will benefit the interested community. Firstly, we explain that the approach of visual matching of coins, universally adopted in all existing published papers on the topic, is not of practical interest because the number of ancient coin types exceeds by far the number of those types which have been imaged, be it in digital form (e.g. online) or otherwise (traditional film, in print, etc.). Rather, we argue that the focus should be on the understanding of the semantic content of coins. Hence, we describe a novel method which uses real-world multimodal input to extract and associate semantic concepts with the correct coin images and then using a novel convolutional neural network learn the appearance of these concepts. Empirical evidence on a real-world and by far the largest data set of ancient coins, we demonstrate highly promising results.

Xingzhi Yue, Neofytos Dimitriou, Ognjen Arandjelovic

Digital pathology (DP) is a new research area which falls under the broad umbrella of health informatics. Owing to its potential for major public health impact, in recent years DP has been attracting much research attention. Nevertheless, a wide breadth of significant conceptual and technical challenges remain, few of them greater than those encountered in the field of oncology. The automatic analysis of digital pathology slides of cancerous tissues is particularly problematic due to the inherent heterogeneity of the disease, extremely large images, amongst numerous others. In this paper we introduce a novel machine learning based framework for the prediction of colorectal cancer outcome from whole digitized haematoxylin & eosin (H&E) stained histopathology slides. Using a real-world data set we demonstrate the effectiveness of the method and present a detailed analysis of its different elements which corroborate its ability to extract and learn salient, discriminative, and clinically meaningful content.

Keylor Daniel Chaves Viquez, Ognjen Arandjelovic, Andrew Blaikie et al.

Access to low cost retinal imaging devices in low and middle income countries is limited, compromising progress in preventing needless blindness. The Arclight is a recently developed low-cost solar powered direct ophthalmoscope which can be attached to the camera of a smartphone to acquire retinal images and video. However, the acquired data is inherently limited by the optics of direct ophthalmoscopy, resulting in a narrow field of view with associated corneal reflections, limiting its usefulness. In this work we describe the first fully automatic method utilizing videos acquired using the Arclight attached to a mobile phone camera to create wider view, higher quality still images comparable with images obtained using much more expensive and bulky dedicated traditional retinal cameras.

Duc-Son Pham, Ognjen Arandjelovic, Svetha Venkatesh

We propose an effective subspace selection scheme as a post-processing step to improve results obtained by sparse subspace clustering (SSC). Our method starts by the computation of stable subspaces using a novel random sampling scheme. Thus constructed preliminary subspaces are used to identify the initially incorrectly clustered data points and then to reassign them to more suitable clusters based on their goodness-of-fit to the preliminary model. To improve the robustness of the algorithm, we use a dominant nearest subspace classification scheme that controls the level of sensitivity against reassignment. We demonstrate that our algorithm is convergent and superior to the direct application of a generic alternative such as principal component analysis. On several popular datasets for motion segmentation and face clustering pervasively used in the sparse subspace clustering literature the proposed method is shown to reduce greatly the incidence of clustering errors while introducing negligible disturbance to the data points already correctly clustered.

Victor Andrei, Ognjen Arandjelovic

The rapidly expanding corpus of medical research literature presents major challenges in the understanding of previous work, the extraction of maximum information from collected data, and the identification of promising research directions. We present a case for the use of advanced machine learning techniques as an aide in this task and introduce a novel methodology that is shown to be capable of extracting meaningful information from large longitudinal corpora, and of tracking complex temporal changes within it.

Reza Shoja Ghiass, Ognjen Arandjelovic

Determining the direction in which a person is looking is an important problem in a wide range of HCI applications. In this paper we describe a highly accurate algorithm that performs gaze estimation using an affordable and widely available device such as Kinect. The method we propose starts by performing accurate head pose estimation achieved by fitting a person specific morphable model of the face to depth data. The ordinarily competing requirements of high accuracy and high speed are met concurrently by formulating the fitting objective function as a combination of terms which excel either in accurate or fast fitting, and then by adaptively adjusting their relative contributions throughout fitting. Following pose estimation, pose normalization is done by re-rendering the fitted model as a frontal face. Finally gaze estimates are obtained through regression from the appearance of the eyes in synthetic, normalized images. Using EYEDIAP, the standard public dataset for the evaluation of gaze estimation algorithms from RGB-D data, we demonstrate that our method greatly outperforms the state of the art.

Warren Rieutort-Louis, Ognjen Arandjelovic

Visual recognition and vision based retrieval of objects from large databases are tasks with a wide spectrum of potential applications. In this paper we propose a novel recognition method from video sequences suitable for retrieval from databases acquired in highly unconstrained conditions e.g. using a mobile consumer-level device such as a phone. On the lowest level, we represent each sequence as a 3D mesh of densely packed local appearance descriptors. While image plane geometry is captured implicitly by a large overlap of neighbouring regions from which the descriptors are extracted, 3D information is extracted by means of a descriptor transition table, learnt from a single sequence for each known gallery object. These allow us to connect local descriptors along the 3rd dimension (which corresponds to viewpoint changes), thus resulting in a set of variable length Markov chains for each video. The matching of two sets of such chains is formulated as a statistical hypothesis test, whereby a subset of each is chosen to maximize the likelihood that the corresponding video sequences show the same object. The effectiveness of the proposed algorithm is empirically evaluated on the Amsterdam Library of Object Images and a new highly challenging video data set acquired using a mobile phone. On both data sets our method is shown to be successful in recognition in the presence of background clutter and large viewpoint changes.

Ognjen Arandjelovic

We are interested in identity-based retrieval of face sets from large unlabelled collections acquired in uncontrolled environments. Given a baseline algorithm for measuring the similarity of two face sets, the meta-algorithm introduced in this paper seeks to leverage the structure of the data corpus to make the best use of the available baseline. In particular, we show how partial transitivity of inter-personal similarity can be exploited to improve the retrieval of particularly challenging sets which poorly match the query under the baseline measure. We: (i) describe the use of proxy sets as a means of computing the similarity between two sets, (ii) introduce transitivity meta-features based on the similarity of salient modes of appearance variation between sets, (iii) show how quasi-transitivity can be learnt from such features without any labelling or manual intervention, and (iv) demonstrate the effectiveness of the proposed methodology through experiments on the notoriously challenging YouTube database and two successful baselines from the literature.

Ieva Vasiljeva, Ognjen Arandjelovic

Large amounts of electronic medical records collected by hospitals across the developed world offer unprecedented possibilities for knowledge discovery using computer based data mining and machine learning. Notwithstanding significant research efforts, the use of this data in the prediction of disease development has largely been disappointing. In this paper we examine in detail a recently proposed method which has in preliminary experiments demonstrated highly promising results on real-world data. We scrutinize the authors' claims that the proposed model is scalable and investigate whether the tradeoff between prediction specificity (i.e. the ability of the model to predict a wide number of different ailments) and accuracy (i.e. the ability of the model to make the correct prediction) is practically viable. Our experiments conducted on a data corpus of nearly 3,000,000 admissions support the authors' expectations and demonstrate that the high prediction accuracy is maintained well even when the number of admission types explicitly included in the model is increased to account for 98% of all admissions in the corpus. Thus several promising directions for future work are highlighted.

Adham Beykikhoshk, Ognjen Arandjelovic, Dinh Phung et al.

In this paper we describe a novel framework for the discovery of the topical content of a data corpus, and the tracking of its complex structural changes across the temporal dimension. In contrast to previous work our model does not impose a prior on the rate at which documents are added to the corpus nor does it adopt the Markovian assumption which overly restricts the type of changes that the model can capture. Our key technical contribution is a framework based on (i) discretization of time into epochs, (ii) epoch-wise topic discovery using a hierarchical Dirichlet process-based model, and (iii) a temporal similarity graph which allows for the modelling of complex topic changes: emergence and disappearance, evolution, splitting, and merging. The power of the proposed framework is demonstrated on two medical literature corpora concerned with the autism spectrum disorder (ASD) and the metabolic syndrome (MetS) -- both increasingly important research subjects with significant social and healthcare consequences. In addition to the collected ASD and metabolic syndrome literature corpora which we made freely available, our contribution also includes an extensive empirical analysis of the proposed framework. We describe a detailed and careful examination of the effects that our algorithms's free parameters have on its output, and discuss the significance of the findings both in the context of the practical application of our algorithm as well as in the context of the existing body of work on temporal topic analysis. Our quantitative analysis is followed by several qualitative case studies highly relevant to the current research on ASD and MetS, on which our algorithm is shown to capture well the actual developments in these fields.

Ognjen Arandjelovic

This paper addresses the problem of automated vehicle tracking and recognition from aerial image sequences. Motivated by its successes in the existing literature focus on the use of linear appearance subspaces to describe multi-view object appearance and highlight the challenges involved in their application as a part of a practical system. A working solution which includes steps for data extraction and normalization is described. In experiments on real-world data the proposed methodology achieved promising results with a high correct recognition rate and few, meaningful errors (type II errors whereby genuinely similar targets are sometimes being confused with one another). Directions for future research and possible improvements of the proposed method are discussed.

Ognjen Arandjelovic

The objective of this work is to automatically recognize faces from video sequences in a realistic, unconstrained setup in which illumination conditions are extreme and greatly changing, viewpoint and user motion pattern have a wide variability, and video input is of low quality. At the centre of focus are face appearance manifolds: this thesis presents a significant advance of their understanding and application in the sphere of face recognition. The two main contributions are the Generic Shape-Illumination Manifold recognition algorithm and the Anisotropic Manifold Space clustering. The Generic Shape-Illumination Manifold is evaluated on a large data corpus acquired in real-world conditions and its performance is shown to greatly exceed that of state-of-the-art methods in the literature and the best performing commercial software. Empirical evaluation of the Anisotropic Manifold Space clustering on a popular situation comedy is also described with excellent preliminary results.

Ognjen Arandjelovic, Duc-Son Pham, Svetha Venkatesh

The need to estimate a particular quantile of a distribution is an important problem which frequently arises in many computer vision and signal processing applications. For example, our work was motivated by the requirements of many semi-automatic surveillance analytics systems which detect abnormalities in close-circuit television (CCTV) footage using statistical models of low-level motion features. In this paper we specifically address the problem of estimating the running quantile of a data stream with non-stationary stochasticity when the memory for storing observations is limited. We make several major contributions: (i) we derive an important theoretical result which shows that the change in the quantile of a stream is constrained regardless of the stochastic properties of data, (ii) we describe a set of high-level design goals for an effective estimation algorithm that emerge as a consequence of our theoretical findings, (iii) we introduce a novel algorithm which implements the aforementioned design goals by retaining a sample of data values in a manner adaptive to changes in the distribution of data and progressively narrowing down its focus in the periods of quasi-stationary stochasticity, and (iv) we present a comprehensive evaluation of the proposed algorithm and compare it with the existing methods in the literature on both synthetic data sets and three large `real-world' streams acquired in the course of operation of an existing commercial surveillance system. Our findings convincingly demonstrate that the proposed method is highly successful and vastly outperforms the existing alternatives, especially when the target quantile is high valued and the available buffer capacity severely limited.

Ognjen Arandjelovic, Duc-Son Pham, Svetha Venkatesh

This paper addresses the task of time separated aerial image registration. The ability to solve this problem accurately and reliably is important for a variety of subsequent image understanding applications. The principal challenge lies in the extent and nature of transient appearance variation that a land area can undergo, such as that caused by the change in illumination conditions, seasonal variations, or the occlusion by non-persistent objects (people, cars). Our work introduces several novelties: (i) unlike all previous work on aerial image registration, we approach the problem using a set-based paradigm; (ii) we show how local, pair-wise constraints can be used to enforce a globally good registration using a constraints graph structure; (iii) we show how a simple holistic representation derived from raw aerial images can be used as a basic building block of the constraints graph in a manner which achieves both high registration accuracy and speed. We demonstrate: (i) that the proposed method outperforms the state-of-the-art for pair-wise registration already, achieving greater accuracy and reliability, while at the same time reducing the computational cost of the task; and (ii) that the increase in the number of available images in a set consistently reduces the average registration error.

Ognjen Arandjelovic, Duc-Son Pham, Svetha Venkatesh

Our aim is to estimate the perspective-effected geometric distortion of a scene from a video feed. In contrast to all previous work we wish to achieve this using from low-level, spatio-temporally local motion features used in commercial semi-automatic surveillance systems. We: (i) describe a dense algorithm which uses motion features to estimate the perspective distortion at each image locus and then polls all such local estimates to arrive at the globally best estimate, (ii) present an alternative coarse algorithm which subdivides the image frame into blocks, and uses motion features to derive block-specific motion characteristics and constrain the relationships between these characteristics, with the perspective estimate emerging as a result of a global optimization scheme, and (iii) report the results of an evaluation using nine large sets acquired using existing close-circuit television (CCTV) cameras. Our findings demonstrate that both of the proposed methods are successful, their accuracy matching that of human labelling using complete visual data.

Adham Beykikhoshk, Ognjen Arandjelovic, Dinh Phung et al.

In this paper we describe a novel framework for the discovery of the topical content of a data corpus, and the tracking of its complex structural changes across the temporal dimension. In contrast to previous work our model does not impose a prior on the rate at which documents are added to the corpus nor does it adopt the Markovian assumption which overly restricts the type of changes that the model can capture. Our key technical contribution is a framework based on (i) discretization of time into epochs, (ii) epoch-wise topic discovery using a hierarchical Dirichlet process-based model, and (iii) a temporal similarity graph which allows for the modelling of complex topic changes: emergence and disappearance, evolution, and splitting and merging. The power of the proposed framework is demonstrated on the medical literature corpus concerned with the autism spectrum disorder (ASD) - an increasingly important research subject of significant social and healthcare importance. In addition to the collected ASD literature corpus which we will make freely available, our contributions also include two free online tools we built as aids to ASD researchers. These can be used for semantically meaningful navigation and searching, as well as knowledge discovery from this large and rapidly growing corpus of literature.

Ognjen Arandjelovic

Clinical trial adaptation refers to any adjustment of the trial protocol after the onset of the trial. The main goal is to make the process of introducing new medical interventions to patients more efficient by reducing the cost and the time associated with evaluating their safety and efficacy. The principal question is how should adaptation be performed so as to minimize the chance of distorting the outcome of the trial. We propose a novel method for achieving this. Unlike previous work our approach focuses on trial adaptation by sample size adjustment. We adopt a recently proposed stratification framework based on collected auxiliary data and show that this information together with the primary measured variables can be used to make a probabilistically informed choice of the particular sub-group a sample should be removed from. Experiments on simulated data are used to illustrate the effectiveness of our method and its application in practice.

Reza Shoja Ghiass, Ognjen Arandjelovic, Hakim Bendada et al.

The reduction of the cost of infrared (IR) cameras in recent years has made IR imaging a highly viable modality for face recognition in practice. A particularly attractive advantage of IR-based over conventional, visible spectrum-based face recognition stems from its invariance to visible illumination. In this paper we argue that the main limitation of previous work on face recognition using IR lies in its ad hoc approach to treating different nuisance factors which affect appearance, prohibiting a unified approach that is capable of handling concurrent changes in multiple (or indeed all) major extrinsic sources of variability, which is needed in practice. We describe the first approach that attempts to achieve this - the framework we propose achieves outstanding recognition performance in the presence of variable (i) pose, (ii) facial expression, (iii) physiological state, (iv) partial occlusion due to eye-wear, and (v) quasi-occlusion due to facial hair growth.

Ognjen Arandjelovic

In this paper we address the problem of matching patterns in the so-called verification setting in which a novel, query pattern is verified against a single training pattern: the decision sought is whether the two match (i.e. belong to the same class) or not. Unlike previous work which has universally focused on the development of more discriminative distance functions between patterns, here we consider the equally important and pervasive task of selecting a distance threshold which fits a particular operational requirement - specifically, the target false positive rate (FPR). First, we argue on theoretical grounds that a data-driven approach is inherently ill-conditioned when the desired FPR is low, because by the very nature of the challenge only a small portion of training data affects or is affected by the desired threshold. This leads us to propose a general, statistical model-based method instead. Our approach is based on the interpretation of an inter-pattern distance as implicitly defining a pattern embedding which approximately distributes patterns according to an isotropic multi-variate normal distribution in some space. This interpretation is then used to show that the distribution of training inter-pattern distances is the non-central chi2 distribution, differently parameterized for each class. Thus, to make the class-specific threshold choice we propose a novel analysis-by-synthesis iterative algorithm which estimates the three free parameters of the model (for each class) using task-specific constraints. The validity of the premises of our work and the effectiveness of the proposed method are demonstrated by applying the method to the task of set-based face verification on a large database of pseudo-random head motion videos.

Ognjen Arandjelovic

Linear subspace representations of appearance variation are pervasive in computer vision. This paper addresses the problem of robustly matching such subspaces (computing the similarity between them) when they are used to describe the scope of variations within sets of images of different (possibly greatly so) scales. A naive solution of projecting the low-scale subspace into the high-scale image space is described first and subsequently shown to be inadequate, especially at large scale discrepancies. A successful approach is proposed instead. It consists of (i) an interpolated projection of the low-scale subspace into the high-scale space, which is followed by (ii) a rotation of this initial estimate within the bounds of the imposed ``downsampling constraint''. The optimal rotation is found in the closed-form which best aligns the high-scale reconstruction of the low-scale subspace with the reference it is compared to. The method is evaluated on the problem of matching sets of (i) face appearances under varying illumination and (ii) object appearances under varying viewpoint, using two large data sets. In comparison to the naive matching, the proposed algorithm is shown to greatly increase the separation of between-class and within-class similarities, as well as produce far more meaningful modes of common appearance on which the match score is based.

Reza Shoja Ghiass, Ognjen Arandjelovic, Hakim Bendada et al.

Automatic face recognition is an area with immense practical potential which includes a wide range of commercial and law enforcement applications. Hence it is unsurprising that it continues to be one of the most active research areas of computer vision. Even after over three decades of intense research, the state-of-the-art in face recognition continues to improve, benefitting from advances in a range of different research fields such as image processing, pattern recognition, computer graphics, and physiology. Systems based on visible spectrum images, the most researched face recognition modality, have reached a significant level of maturity with some practical success. However, they continue to face challenges in the presence of illumination, pose and expression changes, as well as facial disguises, all of which can significantly decrease recognition accuracy. Amongst various approaches which have been proposed in an attempt to overcome these limitations, the use of infrared (IR) imaging has emerged as a particularly promising research direction. This paper presents a comprehensive and timely review of the literature on this subject. Our key contributions are: (i) a summary of the inherent properties of infrared imaging which makes this modality promising in the context of face recognition, (ii) a systematic review of the most influential approaches, with a focus on emerging common trends as well as key differences between alternative methodologies, (iii) a description of the main databases of infrared facial images available to the researcher, and lastly (iv) a discussion of the most promising avenues for future research.

Ognjen Arandjelovic

In contrast to most scientific disciplines, sports science research has been characterized by comparatively little effort investment in the development of relevant phenomenological models. Scarcer yet is the application of said models in practice. We present a framework which allows resistance training practitioners to employ a recently proposed neuromuscular model in actual training program design. The first novelty concerns the monitoring aspect of coaching. A method for extracting training performance characteristics from loosely constrained video sequences, effortlessly and with minimal human input, using computer vision is described. The extracted data is subsequently used to fit the underlying neuromuscular model. This is achieved by solving an inverse dynamics problem corresponding to a particular exercise. Lastly, a computer simulation of hypothetical training bouts, using athlete-specific capability parameters, is used to predict the effected adaptation and changes in performance. The software described here allows the practitioner to manipulate hypothetical training parameters and immediately see their effect on predicted adaptation for a specific athlete. Thus, this work presents a holistic view of the monitoring-assessment-adjustment loop.

Ognjen Arandjelovic

This report describes the design, implementation, evaluation and original enhancements to the Live-Wire method for 2D and 3D image segmentation. Live-Wire 2D employs a semi-automatic paradigm; the user is asked to select a few boundary points of the object to segment, to steer the process in the right direction, while the result is displayed in real time. In our implementation segmentation is extended to three dimensions by performing this process on a slice-by-slice basis. User's time and involvement is further reduced by allowing him to specify object contours in planes orthogonal to the slices. If these planes are chosen strategically, Live-Wire 3D can perform 2D segmentation in the plane of each slice automatically. This report also proposes two improvements to the original method, path heating and a new graph edge feature function based on variance of path properties along the boundary. We show that these improvements lead up to a 33% reduction in interaction with the user, and improved delineation in presence of strong interfering edges.

Ognjen Arandjelovic

The problem of 3D object recognition is of immense practical importance, with the last decade witnessing a number of breakthroughs in the state of the art. Most of the previous work has focused on the matching of textured objects using local appearance descriptors extracted around salient image points. The recently proposed bag of boundaries method was the first to address directly the problem of matching smooth objects using boundary features. However, no previous work has attempted to achieve a holistic treatment of the problem by jointly using textural and shape features which is what we describe herein. Due to the complementarity of the two modalities, we fuse the corresponding matching scores and learn their relative weighting in a data specific manner by optimizing discriminative performance on synthetically distorted data. For the textural description of an object we adopt a representation in the form of a histogram of SIFT based visual words. Similarly the apparent shape of an object is represented by a histogram of discretized features capturing local shape. On a large public database of a diverse set of objects, the proposed method is shown to outperform significantly both purely textural and purely shape based approaches for matching across viewpoint variation.

Ognjen Arandjelovic

The k-means algorithm is a partitional clustering method. Over 60 years old, it has been successfully used for a variety of problems. The popularity of k-means is in large part a consequence of its simplicity and efficiency. In this paper we are inspired by these appealing properties of k-means in the development of a clustering algorithm which accepts the notion of "positively" and "negatively" labelled data. The goal is to discover the cluster structure of both positive and negative data in a manner which allows for the discrimination between the two sets. The usefulness of this idea is demonstrated practically on the problem of face recognition, where the task of learning the scope of a person's appearance should be done in a manner which allows this face to be differentiated from others.

Ognjen Arandjelovic

In this paper we address the problem of matching sets of vectors embedded in the same input space. We propose an approach which is motivated by canonical correlation analysis (CCA), a statistical technique which has proven successful in a wide variety of pattern recognition problems. Like CCA when applied to the matching of sets, our extended canonical correlation analysis (E-CCA) aims to extract the most similar modes of variability within two sets. Our first major contribution is the formulation of a principled framework for robust inference of such modes from data in the presence of uncertainty associated with noise and sampling randomness. E-CCA retains the efficiency and closed form computability of CCA, but unlike it, does not possess free parameters which cannot be inferred directly from data (inherent data dimensionality, and the number of canonical correlations used for set similarity computation). Our second major contribution is to show that in contrast to CCA, E-CCA is readily adapted to match sets in a discriminative learning scheme which we call discriminative extended canonical correlation analysis (DE-CCA). Theoretical contributions of this paper are followed by an empirical evaluation of its premises on the task of face recognition from sets of rasterized appearance images. The results demonstrate that our approach, E-CCA, already outperforms both CCA and its quasi-discriminative counterpart constrained CCA (C-CCA), for all values of their free parameters. An even greater improvement is achieved with the discriminative variant, DE-CCA.

Reza Shoja Ghiass, Ognjen Arandjelovic, Hakim Bendada et al.

Over the course of the last decade, infrared (IR) and particularly thermal IR imaging based face recognition has emerged as a promising complement to conventional, visible spectrum based approaches which continue to struggle when applied in the real world. While inherently insensitive to visible spectrum illumination changes, IR images introduce specific challenges of their own, most notably sensitivity to factors which affect facial heat emission patterns, e.g. emotional state, ambient temperature, and alcohol intake. In addition, facial expression and pose changes are more difficult to correct in IR images because they are less rich in high frequency detail which is an important cue for fitting any deformable model. We describe a novel method which addresses these challenges. To normalize for pose and facial expression changes we generate a synthetic frontal image of a face in a canonical, neutral facial expression from an image of the face in an arbitrary pose and facial expression. This is achieved by piecewise affine warping which follows active appearance model (AAM) fitting. This is the first publication which explores the use of an AAM on thermal IR images; we propose a pre-processing step which enhances detail in thermal images, making AAM convergence faster and more accurate. To overcome the problem of thermal IR image sensitivity to the pattern of facial temperature emissions we describe a representation based on reliable anatomical features. In contrast to previous approaches, our representation is not binary; rather, our method accounts for the reliability of the extracted features. This makes the proposed representation much more robust both to pose and scale changes. The effectiveness of the proposed approach is demonstrated on the largest public database of thermal IR images of faces on which it achieved 100% identification, significantly outperforming previous methods.

Reza Shoja Ghiass, Ognjen Arandjelovic, Hakim Bendada et al.

Over the course of the last decade, infrared (IR) and particularly thermal IR imaging based face recognition has emerged as a promising complement to conventional, visible spectrum based approaches which continue to struggle when applied in practice. While inherently insensitive to visible spectrum illumination changes, IR data introduces specific challenges of its own, most notably sensitivity to factors which affect facial heat emission patterns, e.g. emotional state, ambient temperature, and alcohol intake. In addition, facial expression and pose changes are more difficult to correct in IR images because they are less rich in high frequency detail which is an important cue for fitting any deformable model. In this paper we describe a novel method which addresses these major challenges. Specifically, when comparing two thermal IR images of faces, we mutually normalize their poses and facial expressions by using an active appearance model (AAM) to generate synthetic images of the two faces with a neutral facial expression and in the same view (the average of the two input views). This is achieved by piecewise affine warping which follows AAM fitting. A major contribution of our work is the use of an AAM ensemble in which each AAM is specialized to a particular range of poses and a particular region of the thermal IR face space. Combined with the contributions from our previous work which addressed the problem of reliable AAM fitting in the thermal IR spectrum, and the development of a person-specific representation robust to transient changes in the pattern of facial temperature emissions, the proposed ensemble framework accurately matches faces across the full range of yaw from frontal to profile, even in the presence of scale variation (e.g. due to the varying distance of a subject from the camera).

Reza Shoja Ghiass, Ognjen Arandjelovic, Hakim Bendada et al.

Automatic face recognition (AFR) is an area with immense practical potential which includes a wide range of commercial and law enforcement applications, and it continues to be one of the most active research areas of computer vision. Even after over three decades of intense research, the state-of-the-art in AFR continues to improve, benefiting from advances in a range of different fields including image processing, pattern recognition, computer graphics and physiology. However, systems based on visible spectrum images continue to face challenges in the presence of illumination, pose and expression changes, as well as facial disguises, all of which can significantly decrease their accuracy. Amongst various approaches which have been proposed in an attempt to overcome these limitations, the use of infrared (IR) imaging has emerged as a particularly promising research direction. This paper presents a comprehensive and timely review of the literature on this subject.