Ognjen Arandjelovic | University of St Andrews (original) (raw)

Papers by Ognjen Arandjelovic

IEEE Transactions on Circuits and Systems for Video Technology, 2024

To alleviate the heavy annotation burden for training a reliable crowd counting model and thus ma... more 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 shows strong contextual modeling capabilities, which allows it to make robust predictions even when some local details of patches are lost. 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.

The Open Psychology Journal, 2024

In an era of dramatic technological progress, the consequent economic transformations, and an inc... more In an era of dramatic technological progress, the consequent economic transformations, and an increasing need for an adaptable workforce, the importance of education has risen to the forefront of the social discourse. The concurrent increase in the awareness of issues pertaining to social justice and the debate over what this justice entails and how it ought to be effected, feed into the education policy more than ever before. From the nexus of the aforementioned considerations, a concern over the so-called education gap has emerged, with worldwide efforts to close it. I analyse the premises behind such efforts and demonstrate that they are founded upon fundamentally flawed ideas. I show that in a society in which education is delivered equitably, education gaps emerge naturally as a consequence of differentiation due to talents, the tendency for matched mate selection, and the heritability of intellectual traits. Hence, I issue a call for a refocusing of efforts from the ill-founded idea of closing the education gap, to the understanding of the magnitude of its unfair contributions, as well as to those social aspects which can modulate it in accordance to what a society deems fair according to its values.

Diagnostics, 2024

Amongst other benefits conferred by the shift from traditional to digital pathology is the potent... more Amongst other benefits conferred by the shift from traditional to digital pathology is the potential to use machine learning for diagnosis, prognosis, and personalization. A major challenge in the realization of this potential emerges from the extremely large size of digitized images, often in excess of 100, 000 × 100, 000 pixels. In this paper we tackle this challenge head on, diverging from the existing approaches in the literature which rely on the splitting of the original images into small patches, and introduce magnifying networks (MagNets). Using an attention mechanism MagNets identify the regions of the gigapixel image that benefit from an analysis on a finer scale. This process is repeated, resulting in an attention-driven coarse-to-fine analysis of only a small portion of the information contained in the original whole slide images. Importantly, this is achieved using minimal ground truth annotation, namely using only global, slide-level labels. Our results on the publicly available Camelyon16 and Camelyon17 datasets demonstrate the effectiveness of MagNets, as well as the proposed optimization framework, on the task of whole slide image classification. Importantly, MagNets process at least 5 times fewer patches from each slide than any of the existing end-to-end approaches.

BioMedInformatics, 2024

Background: In recent years, there has been increasing research in the applications of Artificial... more Background: In recent years, there has been increasing research in the applications of Artificial Intelligence in the medical industry. Digital pathology has seen great success in introducing the use of technology in the digitisation and analysis of pathology slides to ease the burden of work on pathologists. Digitised pathology slides, otherwise known as whole slide images, can be analysed by pathologists with the same methods used to analyse traditional glass slides. Methods: The digitisation of pathology slides has also led to the possibility of using these whole slide images to train machine learning models to detect tumours. Patch-based methods are common in the analysis of whole slide images as these images are too large to be processed using normal machine learning methods. However, there is little work exploring the effect that the size of the patches has on the analysis. A patch-based whole slide image analysis method was implemented and then used to evaluate and compare the accuracy of the analysis using patches of different sizes. In addition, two different patch sampling methods are used to test if the optimal patch size is the same for both methods, as well as a downsampling method where whole slide images of low resolution images are used to train an analysis model. Results: It was discovered that the most successful method uses a patch size of 256 × 256 pixels with the informed sampling method, using the location of tumour regions to sample a balanced dataset. Conclusion: Future work on batch-based analysis of whole slide images in pathology should take into account our findings when designing new models.

AISTATS, 2024

In this paper, we tackle the challenge of white-box false positive adversarial attacks on contras... more 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 whitebox 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 whitebox false positive attacks compared to other white-box attack methods.

Journal of Evaluation in Clinical Practice, 2024

Background: Despite the at least decades long record of philosophical recognition and interest, t... more Background: Despite the at least decades long record of philosophical recognition and interest, the intricacy of the deceptively familiar appearing concepts of `disease', `disorder', `disability', etc., has only recently begun showing itself with clarity in the popular discourse wherein its newly emerging prominence stems from the liberties and restrictions contingent upon it. Whether a person is deemed to be afflicted by a disease or a disorder governs their ability to access health care, be it free at the point of use or provided by an insurer; it also influences the treatment of individuals by the judicial system and employers; it even affects one's own perception of self. Aims: All existing philosophical definitions of disease struggle with coherency, causing much confusion and strife, and leading to inconsistencies in real-world practice. Hence, there is a real need for an alternative. Materials & Methods: In the present article I analyse the variety of contemporary views of disease, showing them all to be inadequate and lacking in firm philosophical foundations, and failing to meet the desideratum of patient-driven care. Results: Illuminated by the insights emanating from the said analysis, I introduce a novel approach with firm ethical foundations, which foundations are rooted in sentience, that is the subjective experience of sentient beings. Discussion: I argue that the notion of disease is at best superfluous, and likely even harmful in the provision of compassionate and patient-centred care. Conclusion: Using a series of presently contentious cases illustrate the power of the proposed framework which is capable of providing actionable and humane solutions to problems that leave the current theories confounded.

National Conference on Artificial Intelligence, Jul 14, 2013

IEEE Transactions on Affective Computing, Oct 1, 2022

International Joint Conference on Artificial Intelligence, Jul 9, 2016

Pattern Recognition, Sep 1, 2022

arXiv (Cornell University), Apr 5, 2016

arXiv (Cornell University), Mar 6, 2019

Lecture Notes in Computer Science, 2014

In this paper we describe a novel algorithm for head pose estimation from low-quality RGB-D data ... more In this paper we describe a novel algorithm for head pose estimation from low-quality RGB-D data acquired using a consumer-level device such as Microsoft Kinect. We focus our attention on the well-known challenges in the processing of depth point-clouds which include spurious data, noise, and missing data caused by occlusion. Our algorithm performs pose estimation by fitting a 3D morphable model which explicitly includes pose parameters. Several important novelties are described. (i) We propose a method for automatic removal of the majority of spurious depth data which uses facial feature detection in the associated RGB image. By back-projecting the corresponding image loci and intersecting them with the 3D point-cloud we construct the facial features plane used to crop the point-cloud. (ii) Both high convergence speed and high fitting accuracy are achieved by formulating the fitting objective function to include both point-to-point and point-to-plane point-cloud matching terms. (iii) The effect of misleading point-cloud matches caused by noisy or missing data is reduced by using the Tukey biweight function as a robust statistic and by employing a re-weighting scheme for different terms in the fitting objective function. (iv) Lastly, the proposed algorithm is evaluated on the standard benchmark Biwi Kinect Head Pose Database on which it is shown to outperform substantially the current state-of-the-art, achieving more than a 20-fold reduction in error estimates of all three Euler angles i.e. yaw, pitch, and roll. A thorough analysis of the results is used both to gain full insight into the behaviour of the described algorithm as well as to highlight important methodological issues which future authors should consider in the evaluation of pose estimation algorithms.

PLOS ONE

In this study we use artificial intelligence (AI) to categorise endometrial biopsy whole slide im... more In this study we use artificial intelligence (AI) to categorise endometrial biopsy whole slide images (WSI) from digital pathology as either “malignant”, “other or benign” or “insufficient”. An endometrial biopsy is a key step in diagnosis of endometrial cancer, biopsies are viewed and diagnosed by pathologists. Pathology is increasingly digitised, with slides viewed as images on screens rather than through the lens of a microscope. The availability of these images is driving automation via the application of AI. A model that classifies slides in the manner proposed would allow prioritisation of these slides for pathologist review and hence reduce time to diagnosis for patients with cancer. Previous studies using AI on endometrial biopsies have examined slightly different tasks, for example using images alongside genomic data to differentiate between cancer subtypes. We took 2909 slides with “malignant” and “other or benign” areas annotated by pathologists. A fully supervised convol...

Journal of Imaging

Ancient numismatics, the study of ancient coins, has in recent years become an attractive domain ... more Ancient numismatics, the study of ancient coins, has in recent years become an attractive domain for the application of computer vision and machine learning. Though rich in research problems, the predominant focus in this area to date has been on the task of attributing a coin from an image, that is of identifying its issue. This may be considered the cardinal problem in the field and it continues to challenge automatic methods. In the present paper, we address a number of limitations of previous work. Firstly, the existing methods approach the problem as a classification task. As such, they are unable to deal with classes with no or few exemplars (which would be most, given over 50,000 issues of Roman Imperial coins alone), and require retraining when exemplars of a new class become available. Hence, rather than seeking to learn a representation that distinguishes a particular class from all the others, herein we seek a representation that is overall best at distinguishing classes ...

Data

In recent years, there has been an increased effort to digitise whole-slide images of cancer tiss... more In recent years, there has been an increased effort to digitise whole-slide images of cancer tissue. This effort has opened up a range of new avenues for the application of deep learning in oncology. One such avenue is virtual staining, where a deep learning model is tasked with reproducing the appearance of stained tissue sections, conditioned on a different, often times less expensive, input stain. However, data to train such models in a supervised manner where the input and output stains are aligned on the same tissue sections are scarce. In this work, we introduce a dataset of ten whole-slide images of clear cell renal cell carcinoma tissue sections counterstained with Hoechst 33342, CD3, and CD8 using multiple immunofluorescence. We also provide a set of over 600,000 patches of size 256 × 256 pixels extracted from these images together with cell segmentation masks in a format amenable to training deep learning models. It is our hope that this dataset will be used to further the...

IEEE Transactions on Circuits and Systems for Video Technology, 2024

To alleviate the heavy annotation burden for training a reliable crowd counting model and thus ma... more 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 shows strong contextual modeling capabilities, which allows it to make robust predictions even when some local details of patches are lost. 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.

The Open Psychology Journal, 2024

In an era of dramatic technological progress, the consequent economic transformations, and an inc... more In an era of dramatic technological progress, the consequent economic transformations, and an increasing need for an adaptable workforce, the importance of education has risen to the forefront of the social discourse. The concurrent increase in the awareness of issues pertaining to social justice and the debate over what this justice entails and how it ought to be effected, feed into the education policy more than ever before. From the nexus of the aforementioned considerations, a concern over the so-called education gap has emerged, with worldwide efforts to close it. I analyse the premises behind such efforts and demonstrate that they are founded upon fundamentally flawed ideas. I show that in a society in which education is delivered equitably, education gaps emerge naturally as a consequence of differentiation due to talents, the tendency for matched mate selection, and the heritability of intellectual traits. Hence, I issue a call for a refocusing of efforts from the ill-founded idea of closing the education gap, to the understanding of the magnitude of its unfair contributions, as well as to those social aspects which can modulate it in accordance to what a society deems fair according to its values.

Diagnostics, 2024

Amongst other benefits conferred by the shift from traditional to digital pathology is the potent... more Amongst other benefits conferred by the shift from traditional to digital pathology is the potential to use machine learning for diagnosis, prognosis, and personalization. A major challenge in the realization of this potential emerges from the extremely large size of digitized images, often in excess of 100, 000 × 100, 000 pixels. In this paper we tackle this challenge head on, diverging from the existing approaches in the literature which rely on the splitting of the original images into small patches, and introduce magnifying networks (MagNets). Using an attention mechanism MagNets identify the regions of the gigapixel image that benefit from an analysis on a finer scale. This process is repeated, resulting in an attention-driven coarse-to-fine analysis of only a small portion of the information contained in the original whole slide images. Importantly, this is achieved using minimal ground truth annotation, namely using only global, slide-level labels. Our results on the publicly available Camelyon16 and Camelyon17 datasets demonstrate the effectiveness of MagNets, as well as the proposed optimization framework, on the task of whole slide image classification. Importantly, MagNets process at least 5 times fewer patches from each slide than any of the existing end-to-end approaches.

BioMedInformatics, 2024

Background: In recent years, there has been increasing research in the applications of Artificial... more Background: In recent years, there has been increasing research in the applications of Artificial Intelligence in the medical industry. Digital pathology has seen great success in introducing the use of technology in the digitisation and analysis of pathology slides to ease the burden of work on pathologists. Digitised pathology slides, otherwise known as whole slide images, can be analysed by pathologists with the same methods used to analyse traditional glass slides. Methods: The digitisation of pathology slides has also led to the possibility of using these whole slide images to train machine learning models to detect tumours. Patch-based methods are common in the analysis of whole slide images as these images are too large to be processed using normal machine learning methods. However, there is little work exploring the effect that the size of the patches has on the analysis. A patch-based whole slide image analysis method was implemented and then used to evaluate and compare the accuracy of the analysis using patches of different sizes. In addition, two different patch sampling methods are used to test if the optimal patch size is the same for both methods, as well as a downsampling method where whole slide images of low resolution images are used to train an analysis model. Results: It was discovered that the most successful method uses a patch size of 256 × 256 pixels with the informed sampling method, using the location of tumour regions to sample a balanced dataset. Conclusion: Future work on batch-based analysis of whole slide images in pathology should take into account our findings when designing new models.

AISTATS, 2024

In this paper, we tackle the challenge of white-box false positive adversarial attacks on contras... more 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 whitebox 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 whitebox false positive attacks compared to other white-box attack methods.

Journal of Evaluation in Clinical Practice, 2024

Background: Despite the at least decades long record of philosophical recognition and interest, t... more Background: Despite the at least decades long record of philosophical recognition and interest, the intricacy of the deceptively familiar appearing concepts of `disease', `disorder', `disability', etc., has only recently begun showing itself with clarity in the popular discourse wherein its newly emerging prominence stems from the liberties and restrictions contingent upon it. Whether a person is deemed to be afflicted by a disease or a disorder governs their ability to access health care, be it free at the point of use or provided by an insurer; it also influences the treatment of individuals by the judicial system and employers; it even affects one's own perception of self. Aims: All existing philosophical definitions of disease struggle with coherency, causing much confusion and strife, and leading to inconsistencies in real-world practice. Hence, there is a real need for an alternative. Materials & Methods: In the present article I analyse the variety of contemporary views of disease, showing them all to be inadequate and lacking in firm philosophical foundations, and failing to meet the desideratum of patient-driven care. Results: Illuminated by the insights emanating from the said analysis, I introduce a novel approach with firm ethical foundations, which foundations are rooted in sentience, that is the subjective experience of sentient beings. Discussion: I argue that the notion of disease is at best superfluous, and likely even harmful in the provision of compassionate and patient-centred care. Conclusion: Using a series of presently contentious cases illustrate the power of the proposed framework which is capable of providing actionable and humane solutions to problems that leave the current theories confounded.

National Conference on Artificial Intelligence, Jul 14, 2013

IEEE Transactions on Affective Computing, Oct 1, 2022

International Joint Conference on Artificial Intelligence, Jul 9, 2016

Pattern Recognition, Sep 1, 2022

arXiv (Cornell University), Apr 5, 2016

arXiv (Cornell University), Mar 6, 2019

Lecture Notes in Computer Science, 2014

In this paper we describe a novel algorithm for head pose estimation from low-quality RGB-D data ... more In this paper we describe a novel algorithm for head pose estimation from low-quality RGB-D data acquired using a consumer-level device such as Microsoft Kinect. We focus our attention on the well-known challenges in the processing of depth point-clouds which include spurious data, noise, and missing data caused by occlusion. Our algorithm performs pose estimation by fitting a 3D morphable model which explicitly includes pose parameters. Several important novelties are described. (i) We propose a method for automatic removal of the majority of spurious depth data which uses facial feature detection in the associated RGB image. By back-projecting the corresponding image loci and intersecting them with the 3D point-cloud we construct the facial features plane used to crop the point-cloud. (ii) Both high convergence speed and high fitting accuracy are achieved by formulating the fitting objective function to include both point-to-point and point-to-plane point-cloud matching terms. (iii) The effect of misleading point-cloud matches caused by noisy or missing data is reduced by using the Tukey biweight function as a robust statistic and by employing a re-weighting scheme for different terms in the fitting objective function. (iv) Lastly, the proposed algorithm is evaluated on the standard benchmark Biwi Kinect Head Pose Database on which it is shown to outperform substantially the current state-of-the-art, achieving more than a 20-fold reduction in error estimates of all three Euler angles i.e. yaw, pitch, and roll. A thorough analysis of the results is used both to gain full insight into the behaviour of the described algorithm as well as to highlight important methodological issues which future authors should consider in the evaluation of pose estimation algorithms.

PLOS ONE

In this study we use artificial intelligence (AI) to categorise endometrial biopsy whole slide im... more In this study we use artificial intelligence (AI) to categorise endometrial biopsy whole slide images (WSI) from digital pathology as either “malignant”, “other or benign” or “insufficient”. An endometrial biopsy is a key step in diagnosis of endometrial cancer, biopsies are viewed and diagnosed by pathologists. Pathology is increasingly digitised, with slides viewed as images on screens rather than through the lens of a microscope. The availability of these images is driving automation via the application of AI. A model that classifies slides in the manner proposed would allow prioritisation of these slides for pathologist review and hence reduce time to diagnosis for patients with cancer. Previous studies using AI on endometrial biopsies have examined slightly different tasks, for example using images alongside genomic data to differentiate between cancer subtypes. We took 2909 slides with “malignant” and “other or benign” areas annotated by pathologists. A fully supervised convol...

Journal of Imaging

Ancient numismatics, the study of ancient coins, has in recent years become an attractive domain ... more Ancient numismatics, the study of ancient coins, has in recent years become an attractive domain for the application of computer vision and machine learning. Though rich in research problems, the predominant focus in this area to date has been on the task of attributing a coin from an image, that is of identifying its issue. This may be considered the cardinal problem in the field and it continues to challenge automatic methods. In the present paper, we address a number of limitations of previous work. Firstly, the existing methods approach the problem as a classification task. As such, they are unable to deal with classes with no or few exemplars (which would be most, given over 50,000 issues of Roman Imperial coins alone), and require retraining when exemplars of a new class become available. Hence, rather than seeking to learn a representation that distinguishes a particular class from all the others, herein we seek a representation that is overall best at distinguishing classes ...

Data

In recent years, there has been an increased effort to digitise whole-slide images of cancer tiss... more In recent years, there has been an increased effort to digitise whole-slide images of cancer tissue. This effort has opened up a range of new avenues for the application of deep learning in oncology. One such avenue is virtual staining, where a deep learning model is tasked with reproducing the appearance of stained tissue sections, conditioned on a different, often times less expensive, input stain. However, data to train such models in a supervised manner where the input and output stains are aligned on the same tissue sections are scarce. In this work, we introduce a dataset of ten whole-slide images of clear cell renal cell carcinoma tissue sections counterstained with Hoechst 33342, CD3, and CD8 using multiple immunofluorescence. We also provide a set of over 600,000 patches of size 256 × 256 pixels extracted from these images together with cell segmentation masks in a format amenable to training deep learning models. It is our hope that this dataset will be used to further the...

Identifying the configuration of chess pieces from an image of a chessboard is a problem in compu... more Identifying the configuration of chess pieces from an image of a chessboard is a problem in computer vision that has not yet been solved accurately. However, it is important for helping amateur chess players improve their games by facilitating automatic computer analysis without the overhead of manually entering the pieces. Current approaches are limited by the lack of large datasets and are not designed to adapt to unseen chess sets. This paper puts forth a new dataset synthesised from a 3D model that is an order of magnitude larger than existing ones. Trained on this dataset, a novel end-to-end chess recognition system is presented that combines traditional computer vision techniques with deep learning. It localises the chessboard using a RANSAC-based algorithm that computes a projective transformation of the board onto a regular grid. Using two convolutional neural networks, it then predicts an occupancy mask for the squares in the warped image and finally classifies the pieces. The described system achieves an error rate of 0.23% per square on the test set, 28 times better than the current state of the art. Further, a few-shot transfer learning approach is developed that is able to adapt the inference system to a previously unseen chess set using just two photos of the starting position, obtaining a per-square accuracy of 99.83% on images of that new chess set. The dataset is released publicly; code and trained models are available at this URL.