<> "The repository administrator has not yet configured an RDF license."^^ . <> . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells"^^ . "With the continuous advancements in microscopy techniques such as improved image quality,\r\nfaster acquisition and reduced photo-toxicity, the amount of data recorded in the life sciences\r\nis rapidly growing. Clearly, the size of the data renders manual analysis intractable, calling\r\nfor automated cell tracking methods. Cell tracking – in contrast to other tracking scenarios\r\n– exhibits several difficulties: low signal to noise ratio in the images, high cell density and\r\nsometimes cell clusters, radical morphology changes, but most importantly cells divide – which\r\nis often the focus of the experiment. These peculiarities have been targeted by tracking-byassignment\r\nmethods that first extract a set of detection hypotheses and then track those over\r\ntime. Improving the general quality of these cell tracking methods is difficult, because every cell\r\ntype, surrounding medium, and microscopy setting leads to recordings with specific properties\r\nand problems. This unfortunately implies that automated approaches will not become perfect\r\nany time soon but manual proof reading by experts will remain necessary for the time being.\r\nIn this thesis we focus on two different aspects, firstly on scaling previous and developing new\r\nsolvers to deal with longer videos and more cells, and secondly on developing a specialized\r\npipeline for detecting and tracking tuberculosis bacteria.\r\nThe most powerful tracking-by-assignment methods are formulated as probabilistic graphical\r\nmodels and solved as integer linear programs. Because those integer linear programs are in\r\ngeneral NP-hard, increasing the problem size will lead to an explosion of computational cost.\r\nWe begin by reformulating one of these models in terms of a constrained network flow, and\r\nshow that it can be solved more efficiently. Building on the successful application of network\r\nflow algorithms in the pedestrian tracking literature, we develop a heuristic to integrate constraints\r\n– here for divisions – into such a network flow method. This allows us to obtain high\r\nquality approximations to the tracking solution while providing a polynomial runtime guarantee.\r\nOur experiments confirm this much better scaling behavior to larger problems. However, this\r\napproach is single threaded and does not utilize available resources of multi-core machines yet.\r\nTo parallelize the tracking problem we present a simple yet effective way of splitting long videos\r\ninto intervals that can be tracked independently, followed by a sparse global stitching step that\r\nresolves disagreements at the cuts. Going one step further, we propose a microservices based\r\nsoftware design for ilastik that allows to distribute all required computation for segmentation,\r\nobject feature extraction, object classification and tracking across the nodes of a cluster or in the\r\ncloud.\r\nFinally, we discuss the use case of detecting and tracking tuberculosis bacteria in more\r\ndetail, because no satisfying automated method to this important problem existed before. One\r\npeculiarity of these elongated cells is that they build dense clusters in which it is hard to outline individuals. To cope with that we employ a tracking-by-assignment model that allows competing\r\ndetection hypotheses and selects the best set of detections while considering the temporal context\r\nduring tracking. To obtain these hypotheses, we develop a novel algorithm that finds diverseM-\r\nbest solutions of tree-shaped graphical models by dynamic programming. First experiments\r\nwith the pipeline indicate that it can greatly reduce the required amount of human intervention\r\nfor analyzing tuberculosis treatment."^^ . "2017" . . . . . . . "Carsten"^^ . "Haubold"^^ . "Carsten Haubold"^^ . . . . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (PDF)"^^ . . . "haubold-2017-phd_thesis.pdf"^^ . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (Other)"^^ . . . . . . "indexcodes.txt"^^ . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (Other)"^^ . . . . . . "lightbox.jpg"^^ . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (Other)"^^ . . . . . . "preview.jpg"^^ . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (Other)"^^ . . . . . . "medium.jpg"^^ . . . "Scalable Inference for Multi-Target Tracking\r\nof Proliferating Cells (Other)"^^ . . . . . . "small.jpg"^^ . . "HTML Summary of #23316 \n\nScalable Inference for Multi-Target Tracking \nof Proliferating Cells\n\n" . "text/html" . . . "004 Informatik"@de . "004 Data processing Computer science"@en . .