35 research outputs found
What’s Lost in Campaigns Against Sex Trafficking?
No full textIn recent years, media and public attention about sex trafficking has grown enormously. Anti-trafficking activism has focused primarily on \u27rescuing¹ women forced into prostitution, (often described as \u27modern-day slavery) as well as increasing criminal enforcement of laws against men who profit from prostitution. What impact has this current wave of activism had on long-standing feminist approaches to ending gender-based violence? What has been the impact when anti-trafficking laws are aggressively enforced? And how does use of the term slavery here relate to the broader histories of white supremacy and racial domination?
This talk engages these questions, and considers the race and gender - based effects of anti-trafficking efforts and their implications for the future of feminist organizing around labor rights and sexual violence.
About the Lecturers:
Priscilla Yamin is an Associate Professor of Political Science at the University of Oregon. Her research and teaching focus on gender, sexuality and race politics in the United States. She recently published a book on marriage politics, American Marriage: A Political Institution.
Daniel Martinez HoSang is an Associate Professor at the University of Oregon with a joint appointment in the Department of Ethnic Studies and the Department of Political Science. He is the author of Racial Propositions: Ballot Initiatives and the Making of Postwar California and co-editor of Racial Formation in the 21st Century. He writes and teaches about racial politics and racial justice, social movements, labor and gender
Analyse und Verbesserung des visuellen Objektdetektionsverfahrens
No full textVisual object detection has seen substantial improvements during the last years due to the possibilities enabled by deep learning. While research on image classification provides continuous progress on how to learn image representations and classifiers jointly, object detection research focuses on identifying how to properly use deep learning technology to effectively localise objects. In this thesis, we analyse and improve different aspects of the commonly used detection pipeline. We analyse ten years of research on pedestrian detection and find that improvement of feature representations was the driving factor. Motivated by this finding, we adapt an end-to-end learned detector architecture from general object detection to pedestrian detection. Our deep network outperforms all previous neural networks for pedestrian detection by a large margin, even without using additional training data. After substantial improvements on pedestrian detection in recent years, we investigate the gap between human performance and state-of-the-art pedestrian detectors. We find that pedestrian detectors still have a long way to go before they reach human performance, and we diagnose failure modes of several top performing detectors, giving direction to future research. As a side-effect we publish new, better localised annotations for the Caltech pedestrian benchmark. We analyse detection proposals as a preprocessing step for object detectors. We establish different metrics and compare a wide range of methods according to these metrics. By examining the relationship between localisation of proposals and final object detection performance, we define and experimentally verify a metric that can be used as a proxy for detector performance. Furthermore, we address a structural weakness of virtually all object detection pipelines: non-maximum suppression. We analyse why it is necessary and what the shortcomings of the most common approach are. To address these problems, we present work to overcome these shortcomings and to replace typical non-maximum suppression with a learnable alternative. The introduced paradigm paves the way to true end-to-end learning of object detectors without any post-processing. In summary, this thesis provides analyses of recent pedestrian detectors and detection proposals, improves pedestrian detection by employing deep neural networks, and presents a viable alternative to traditional non-maximum suppression.Die visuelle Objektdetektion erfuhr in den letzten Jahren durch die Möglichkeiten von Deep Learning erhebliche qualitative Verbesserungen. Während durch die Forschung zur Bildklassifizierung kontinuierliche Fortschritte darin erzielt werden, wie Merkmalsrepräsentation und Klassifikatoren gemeinsam gelernt werden, konzentriert sich die Forschung zur Objektdetektion darauf, wie Deep Learning verwendet werden kann, um Objekte schnell und genau zu lokalisieren. In dieser Arbeit analysieren und verbessern wir verschiedene Aspekte des häufig verwendeten Objektdetektions-Prozesses. Wir analysieren den Fortschritt von zehn Jahren Forschung an Fußgängererkennung und finden heraus, dass die Verbesserung von Merkmalsrepräsentationen den Schlüsselfaktor darstellt. Durch diese Erkenntnis motiviert, adaptieren wir ein tiefes neuronales Netzwerk zur allgemeinen Objekterkennung, das Merkmalsrepräsentation und Klassifikatoren gemeinsam lernt, für die Fußgängererkennung. Unser Netzwerk übertrifft alle bisherigen neuronalen Netze für die Fußgängererkennung bei Weitem, sogar wenn keine zusätzlichen Trainingsdaten verwendet werden. Nach signifikanten Verbesserungen der Fußgängererkennung in den letzten Jahren untersuchen wir den qualitativen Unterschied zwischen menschlicher Leistung und Ergebnissen von Fußgängerdetektoren auf dem neuesten Stand der Technik. Unsere Experimente zeigen, dass Fußgängerdetektoren noch einen langen Weg vor sich haben um menschliche Qualität zu erreichen. Wir untersuchen Fehler von mehreren starken Fußgängerdetektoren und charakterisieren häufige Fehlerquellen. Ein Nebenprodukt dieser Arbeit sind neue und besser lokalisierte Annotationen für den Caltech Fußgängerdetektions-Benchmark. Wir analysieren Erkennungsvorschläge (detection proposals) als Vorverarbeitungsschritt für Objektdetektion. Wir definieren verschiedene Metriken und vergleichen eine breite Palette von Methoden nach diesen Metriken. Durch die Untersuchung der Beziehung zwischen der Lokalisierung von Erkennungsvorschlägen und der endgültigen Objektdetektionsleistung definieren und verifizieren wir experimentell eine Metrik, die als Stellvertreter für die Detektorleistung verwendet werden kann. Darüber hinaus behandeln wir eine strukturelle Schwäche von praktisch allen Objekterkennungs-Prozessen: Unterdrückung nicht-maximaler Detektionen. Wir analysieren, warum dieser Schritt notwendig ist und was die Unzulänglichkeiten des gebräuchlichen Ansatzes sind. Um diese Probleme zu lösen, stellen wir Forschung vor, die diese Mängel überwindet und die die typische Unterdrückung durch eine erlernbare Alternative ersetzt. Das vorgestellte Paradigma ebnet den Weg zu echtem End-to-End-Lernen'' von Objektdetektoren, die keine weitere Nachbearbeitung benötigen. Zusammenfassend stellt diese Dissertation Analysen der jüngsten Fußgänger-Detektoren und Erkennungsvorschläge vor, verbessert die Fußgängererkennung durch den Einsatz tiefer neuronaler Netze und präsentiert eine tragfähige Alternative zur herkömmlichen Unterdrückung nicht-maximaler Detektionen
A Convnet for Non-maximum Suppression
No full textNon-maximum suppression (NMS) is used in virtually all state-of-the-art object detection pipelines. While essential object detection ingredients such as features, classifiers, and proposal methods have been extensively researched surprisingly little work has aimed to systematically address NMS. The de-facto standard for NMS is based on greedy clustering with a fixed distance threshold, which forces to trade-off recall versus precision. We propose a convnet designed to perform NMS of a given set of detections. We report experiments on a synthetic setup, and results on crowded pedestrian detection scenes. Our approach overcomes the intrinsic limitations of greedy NMS, obtaining better recall and precision
An Evaluation of Two Automatic Landmark Building Discovery Algorithms for City Reconstruction
No full textHow good are detection proposals, really?
No full textCurrent top performing Pascal VOC object detectors employ detection proposals to guide the search for objects thereby avoiding exhaustive sliding window search across images. Despite the popularity of detection proposals, it is unclear which trade‐offs are made when using them during object detection. We provide an in depth analysis of ten object proposal methods along with four baselines regarding ground truth annotation recall (on Pascal VOC 2007 and ImageNet 2013), repeatability, and impact on DPM detector performance. Our findings show common weaknesses of existing methods, and provide insights to choose the most adequate method for different settings
Dietary cholesterol promotes repair of demyelinated lesions in the adult brain
No full textMultiple Sclerosis (MS) is an inflammatory demyelinating disorder in which remyelination failure contributes to persistent disability. Cholesterol is rate-limiting for myelin biogenesis in the developing CNS; however, whether cholesterol insufficiency contributes to remyelination failure in MS, is unclear. Here, we show the relationship between cholesterol, myelination and neurological parameters in mouse models of demyelination and remyelination. In the cuprizone model, acute disease reduces serum cholesterol levels that can be restored by dietary cholesterol. Concomitant with blood-brain barrier impairment, supplemented cholesterol directly supports oligodendrocyte precursor proliferation and differentiation, and restores the balance of growth factors, creating a permissive environment for repair. This leads to attenuated axon damage, enhanced remyelination and improved motor learning. Remarkably, in experimental autoimmune encephalomyelitis, cholesterol supplementation does not exacerbate disease expression. These findings emphasize the safety of dietary cholesterol in inflammatory diseases and point to a previously unrecognized role of cholesterol in promoting repair after demyelinating episodes
An Evaluation of Two Automatic Landmark Building Discovery Algorithms for City Reconstruction
No full textAn Evaluation of Two Automatic Landmark Building Discovery Algorithms for City Reconstruction
No full text
