We present a novel approach to Bayesian inference and general Bayesian computation that is defined through a sequential decision loop. Our method defines a recursive partitioning of the sample space. It neither relies on gradients nor requires any problem-specific tuning, and is asymptotically exact for any density function with a bounded domain. The output is an approximation to the whole density function including the normalisation constant, via partitions organised in efficient data structures. Such approximations may be used for evidence estimation or fast posterior sampling, but also as building blocks to treat a larger class of estimation problems. The algorithm shows competitive performance to recent state-of-the-art methods on synthetic and real-world problems including parameter inference for gravitational-wave physics.

Bayesian optimization (BO) methods often rely on the assumption that the objective function is well-behaved, but in practice, this is seldom true for real-world objectives even if noise-free observations can be collected. Common approaches, which try to model the objective as precisely as possible, often fail to make progress by spending too many evaluations modeling irrelevant details. We address this issue by proposing surrogate models that focus on the well-behaved structure in the objective function, which is informative for search, while ignoring detrimental structure that is challenging to model from few observations. First, we demonstrate that surrogate models with appropriate noise distributions can absorb challenging structures in the objective function by treating them as irreducible uncertainty. Secondly, we show that a latent Gaussian process is an excellent surrogate for this purpose, comparing with Gaussian processes with standard noise distributions. We perform numerous experiments on a range of BO benchmarks and find that our approach improves reliability and performance when faced with challenging objective functions.

The objective of this paper is the fully automated visual identification of individual Holstein Friesian cattle from dorsal RGB-D imagery taken in real-world farm environments. Autonomous and non-intrusive cattle identification could provide an essential tool for economically-viable machinised farming analytics, social monitoring, cattle traceability, food production management and more. We contribute a dataset and propose a system that can reliably derive animal identities from top-down stills by first depth-segmenting animals in RGB-D frames, and then extracting a subset of local ASIFT coat descriptors predicted as sufficiently individually distinctive across the species. Predictions are generated by a support vector machine (SVM) using radial basis function (RBF) kernels for predictions based on the ASIFT descriptor structure. We show that learning such a species-specific ID-model is effective, and we demonstrate robustness to poor or complex input image conditions such as more than one cow present, bad depth segmentation, etc. The proposed system yields 97% identification accuracy over testing on approximately 86,000 image pair comparisons covering a herd of 40 individuals from the FriesianCattle2015 Dataset.

We present an application that demonstrates a proof of concept system for automated in-the-field monitoring of disease in wheat crops. Such in-situ applications are required to be robust in the presence of clutter, provide rapid and accurate analysis and are able to operate at scale. We propose a processing pipeline that detects key wheat diseases in cluttered field imagery. First, we describe and evaluate a high dimensional texture descriptor combined with a randomised forest approach for automated primary leaf recognition. Second, we show that a combined nearest neighbour classifier and voting system applied to segmented leaf regions can robustly determine the presence and type of disease. The system has been tested on a real-world database of images of wheat leaves captured in-the-field using a standard smart phone.

This paper presents a system for producing multiperspective panoramas for long scenes from dense collections of images. Multi-perspective panoramas are generated by combining different perspectives, including original and novel perspectives. The latter are rendered by our synthesis algorithm, which uses estimated depth information to remove distortions caused by depth parallax. Then, different perspectives are properly combined without noticeable visual artifacts caused by both color inconsistencies and structural misalignments.

We present a method that is capable of tracking and estimating pose of articulated objects in real-time. This is achieved by using a bottom-up approach to detect instances of the object in each frame, these detections are then linked together using a high-level a priori motion model. Unlike other approaches that rely on appearance, our method is entirely dependent on motion; initial low-level part detection is based on how a region moves as opposed to its appearance. This work is best described as pictorial structures using motion. A sparse cloud of points extracted using a standard feature tracker are used as observational data, this data contains noise that is not Gaussian in nature but systematic due to tracking errors. Using a probabilistic framework we are able to overcome both corrupt and missing data whilst still inferring new poses from a generative model. Our approach requires no manual initialisation and we show results for a number of complex scenes and different classes of articulated object, this demonstrates both the robustness and versatility of the presented technique.

In order to represent large amounts of information in the form of a video key-frame summary, this paper studies narrative grammar of comics, and using its universal and intuitive rules, lays out visual summaries in an efficient and user centered way. The system ranks importance of key-frame sizes in the final layout by balancing the dominant visual representability and discovery of unanticipated content utilizing a specific cost function and an unsupervised robust spectral clustering technique. A final layout is created using an optimization algorithm based on dynamic programming. Algorithm efficiency and robustness are demonstrated by comparing the results with the optimal panelling solutions.

This paper describes a novel approach to detecting walking quadrupeds in unedited wildlife film footage. Variable lighting, moving backgrounds and camouflaged animals make traditional foreground extraction techniques such as optical flow and background subtraction unstable. We track a sparse set of points over a short film clip and interpolate dense flow, using normalized convolution. Principal component analysis (PCA) is applied to a set of dense flows, describing quadruped gait and other movements. The projection coefficients for relevant principal components are analysed as one dimensional time series. Projection coefficient variation reflects changes in the velocity and relative alignment of the components of the foreground object. These coefficients' relative phase differences are used to train a KNN classifier, which segments the training data with 93% success rate. By generating projection coefficients for unseen footage, the system has successfully located examples of quadruped gait previously missed by human observers.

We present a novel approach to motion synthesis. It is shown that by splitting sequences into segments new sequences can be created with a similar look and feel to the original. Copying segments of the original data generates a sequence which maintains detailed characteristics. By modelling each segment using an autoregressive process we can introduce new segments and therefore unseen motions. These statistical models allow a potentially infinite number of new segments to be generated. We show that this system can model complicated nonstationary sequences which a single ARP is unable to do

In this paper we propose a system that recognises gait and quadruped structure from a sparse set of tracked points. In this work the motion information is derived from dynamic wildlife film footage and is consequently extremely complex and noisy. The gait analysis is carried out on footage that contains quadrupeds, usually walking in profile to the camera, however, this part of the system is pose independent and useful for gait detection in general. The dominant motion is assumed to generate by the background and its relationship to the camera motion, enabling its removal as an initial step. Along with frequency analysis, an eigengait model is used as a template to synchronise clusters of points and to established an underlying spatio-temporal structure. Given this synchronised structure, further tracking observations are used to deform the structure to better fit the overall motion. We demonstrate that the use of an eigengait model enables the spatio-temporal localisation of walking animals and assists in overcoming difficulties caused by occlusion, tracking failure and noisy measurements.

We propose a novel method that uses simulated annealing to create an optimal surface mesh by selecting a subset of a 3D point cloud and a triangulation that reliably represents the actual topology of the scene. This method provides a number of advantages: it copes well with noisy data, it produces a simplified mesh, particularly for scenes that contain many planes and, unlike greedy search techniques, it is much more likely to converge to a global minimum.

We describe a system which is designed to assist animators in extracting high-level information from sequences of images. The system is not meant to replace animators, but to be a tool to assist them in creating the first `rough-cut' of a sequence quickly and easily. Using the system, short animations have been created in a very short space of time. We show that the method of principal components analysis followed by a neural network learning phase is capable of motion tracking (even through occlusion), feature-extraction and gait classification. We quantify the results, and demonstrate the system tracking horses, birds and actors in a film. We demonstrate a system that is powerful, flexible and, above all, easy for non-specialists to use.

We describe a system which is designed to assist in extracting high-level information from sets or sequences of images. We show that the method of principal components analysis followed by a neural network learning phase is capable of feature extraction or motion tracking, even through occlusion. Given a minimum amount of user direction for the learning phase, a wide range of features can be automatically extracted. Features discussed in this paper include information associated with human head motions and a birds wings during take-off. We have quantified the results, for instance showing that with only 25 out of 424 frames of hand labelled information a system to track a persons nose can be trained almost as accurately as a human attempting the same task. We demonstrate a system that is powerful, flexible and, above all, easy for nonspecialists to use.

Fourier analysis is often used as a tool to facilitate the extraction of texture information from image data. However, in situations where the texture patch does not entirely fill the region of analysis, information relating to the shape of the patch becomes entwined with its texture content, thus contaminating the Fourier spectrum and corrupting the texture information. We propose the use of a frequency deconvolution algorithm to remove the artefacts introduced by shape components, permitting the Fourier-based analysis of non-rectangular image patches. The algorithm is demostrated on a texture recognition task involving the entire Brodatz album.

The paper describes how neural networks may be used to segment and label objects in images. A self-organising feature map is used for the segmentation phase, and we quantify the quality of the segmentations produced as well as the contribution made by colour and texture features. A multi-layer perception is trained to label the regions produced by the segmentation process. It is shown that 91.1% of the image area is correctly classified into one of eleven categories which include cars, houses, fences, roads, vegetation and sky.

This paper describes recent work on a neural network approach to outdoor scene interpretation. The results of evaluating a range of automatic region-based segmentation techniques based on a new segmentation quality metric are presented. The optimal technique is used to segment images of natural outdoor scenes. A powerful set of features designed for outdoor scene analysis is extracted from regions in the segmented images and used to train a neural network to recognise eleven different classes of objects, including sky, road, building and vegetation. The system is tested on a large number of images which have been hand-labelled to provide ground-truth segmentations and interpretations. This fully automated system achieves an impressive mean classification accuracy of 81.4 % per image on unseen data, and runs in approximately 1 CPU minute on a Sun SPARCstation 20. 1