We tackle the task of semi-supervised video object segmentation, i.e, pixel-level object classification of the images in video sequences using very limited ground truth training data of its corresponding video. Recently introduced online adaptation of convolutional neural networks for video object segmentation (OnAVOS) has achieved good results by pretraining the network, fine-tuning on the first frame and training the network at test time using its approximate prediction as newly obtained ground truth. We propose Flow Adaptive Video Object Segmentation (FAVOS) that refines the generated adaptive ground truth for online updates and utilizes temporal consistency between video frames with the help of optical flow. We validate our approach on the DAVIS Challenge and achieve rank 1 results on the DAVIS 2016 Challenge (single-object segmentation) and competitive scores on both DAVIS 2018 Semi-supervised Challenge and Interactive Challenge (multi-object segmentation). While most models tend to have increasing complexity for the challenging task of video object segmentation, FAVOS provides a simple and efficient pipeline that produces accurate predictions.

The presence of toxic order ow and predatory HFT strategies in a multi-market setting are scarcely researched in the academic world. This thesis studies the toxicity levels of a set of markets by examining unconsolidated quote data and firm specific trade data. A method for deducing the markets toxicity levels is presented along with proxies for toxic order ow, namely: changes in spread and quoted volume, following a trade in a given market. We find both signs of toxicity and different toxicity levels between the markets. However, the results are lacking in statistical significance but they show that this field is of great interest for further research. Also, the methods proposed for deducing the toxicity levels are rudimentary but could serve well as a premise for further development. / Närvaron av toxic order flow och predatoriska HFT-strategier i en flermarknadsmiljö är föga studerat i den akademiska världen. Denna avhandling studerar detta på en uppsättning marknader genom att undersöka okonsoliderad quote data och firma specifika trades, och på så vis ta fram marknadernas toxicity-nivåer. En metod för att fastställa marknadernas toxicity-nivåer presenteras tillsammans med proxys för toxic order flow, mer specifikt: förändringar i spread och quotad volym, efter en handel på en given marknad. Vi finner både tecken på toxicity och olika toxicityniv åer mellan marknaderna. Resultaten saknar dock statistisk signifikans men de visar ändå på att detta område är av stort intresse för ytterligare forskning. De metoder som föreslås för att fastställa toxicity-nivåerna är rudimentära, men kan tjäna som en utgångspunkt för vidare utveckling.

Les mécanismes d’exhumation des roches de haute-pression basse-température (HP-BT) au sein des zones de subduction sont nombreux, complexes et toujours activement débattus. L’étude des zones de subduction fossiles permet alors de mieux comprendre ces mécanismes, montrant notamment que l’exhumation des roches métamorphiques HP-BT est en grande partie accommodée le long de zones de cisaillement ductiles d’échelle crustale. Ce travail de thèse vise à contraindre la géométrie, la cinématique et la durée de l’histoire tectono-métamorphique des roches de HP-BT du complexe de subduction des Schistes Bleus Cycladiques située en Grèce. Un des objectifs est de contraindre, à différentes échelles, le calendrier de la localisation de la déformation au cours de l’exhumation dans le but de mieux comprendre le comportement mécanique des zones de subduction. Trois méthodes principales d’investigation ont été utilisées : i) une étude de terrain permettant de caractériser la géométrie, la cinématique et la distribution de la déformation, montrant notamment une localisation progressive de la déformation vers la base de l’unité des Schistes Bleus Cycladiques au cours de l’exhumation et le long de zones de cisaillement, ii) des outils de pétrologie métamorphique permettant de contraindre l’évolution P-T des roches métamorphiques au sein de la zone de subduction, et iii) des datations ⁴⁰AR/³⁹AR afin de déterminer le calendrier de l’histoire tectono-métamorphique des Schistes Bleus Cycladiques et de la localisation de la déformation au sein de zones de cisaillement d’échelle kilométrique à millimétrique associées à des degrés de rétromorphose variés. Une corrélation nette est clairement observée entre l’intensité de la déformation, le degré de rétromorphose et les âges les plus jeunes. Un des résultats de ce travail est que la préservation à l’affleurement d’éclogites et schistes bleus n’implique pas forcément une exhumation rapide. En effet, nos résultats suggèrent que l’unité des Schistes Bleus Cycladiques a enregistré une histoire d’exhumation longue d’environ 30 Ma. Le taux d’exhumation n’est donc pas le seul paramètre contrôlant le degré de rétromorphose des unités HP-BT, la remontée le long d’un gradient métamorphique froid dans le canal de subduction et la localisation progressive de la déformation ductile au cours de l’exhumation étant également des facteurs majeurs. / Exhumation mechanisms of high-pressure low-temperature (HP-LT) metamorphic rocks in subduction zones are complex and actively discussed. The study of fossilized subduction zones allows a better comprehension of these mechanisms, showing that exhumation of HP-LT rocks is mainly accommodated along crustal-scale ductile shear zones. This study aims at constraining the geometry, the kinematic and the timing of the tectonometamorphic history of the HP-LT Cycladic Blueschist Unit (CBU) cropping out in Greece. A main objective is to constrain the timing of strain localization at different scales during exhumation to better understand the mechanical behaviour of subduction zones. Three principal methods of investigation have been used, including i) a structural fieldwork that allows to characterize the geometry, the kinematic and the distribution of deformations, highlighting progressive strain localization during exhumation toward the base of the CBU and along shear zones, ii) a metamorphic petrology study aiming at determining the P-T evolution of the CBU, and iii) ⁴⁰AR/³⁹AR dating used to constrain the timing of the tectonometamorphic evolution of the CBU and the timing of strain localization within kilometre- to millimetre-scale shear zones showing different degrees of retrogression. We observe an obvious correlation between the intensity of finite deformation, the degree of retrogression and youngest mica ages. A major result of this thesis work is that the preservation of eclogite and blueschist-facies rocks does not necessarily imply fast exhumation rates. Our results instead suggest that the exhumation history of the CBU is relatively long, spanning over ca. 30 Ma. Consequently, it appears that the exhumation rate is not the main parameter controlling the degree of retrogression of HP-LT metamorphic rocks in the CBU compared to progressive strain localization during exhumation along a cold retrograde P-T evolution within the subduction channel.

The ow of a liquid _lm over a rapidly rotating horizontal disk has numerous industrial applications including pharmaceuticals, chemical engineering, bioengineering, etc. The analysis and control of complex uid ows over a rapidly rotating horizontal disk is an important issue in the experimental uid mechanics. The spinning disk reactor exploits the bene_ts of centrifugal force, which produces thin highly sheared _lms due to radial acceleration. The hydrodynamics of the _lm results in excellent uid mixing and high heat or mass transfer rates. This work focuses on developing a novel approach for uid ow tracking and analysis. Speci_cally, the developed algorithm is able to detect the moving waves and compute controlling _lm ow parameters for the uid owing over a rotating disk. The input to this algorithm is an easily acquired non-invasive video data. It is shown that under single light illumination it is possible to track specular portion of the reected light on the moving wave. Hence, the uid wave motion can be tracked and uid ow parameters can be computed. The uid ow parameters include wave velocities, wave inclination angles, and distances between consecutive waves. Once the parameters are computed, their accuracy is analyzed and compared with the solutions of the mathematical uid dynamics models based on the Navier-Stokes equations for the case of a thin _lm. The uid model predicts wave characteristics based on directly measured controlling parameters, such as disk rotation speed and uid ow rate. It is shown that the calculated parameter values approximately coincide with the predicted ones. The average computed parameters were within 5 � 10% of the predicted values. In addition, given recovered uid characteristics and uid ow controlling parameters, full 3D wave description is obtained. That includes 3D wave location, speed, and distance between waves, as well as approximate wave thickness. Next, the developed approach is generalized to model-based recovery of uid ow controlling parameters: the speed of the spinning disk and the initial uid-ow rate. The search in space for model parameters is performed as to minimize the error between the ow characteristics predicted by the uid dynamics model (e.g. distance between waves, wave inclination angles) and parameters recovered from video data. Results demonstrate that the speed of a disk and the ow rate are recovered with high accuracy. When compared to the ground truth available from direct observation, we noted that the controlling parameters were estimated with less than 10% error.

Le but de cette thèse est d'étudier le transport et le dépôt de particules solides dans les écoulements à travers les fractures. Dans un premier temps, l'écoulement monophasique à travers les fractures est étudié afin d'évaluer la validité de la loi cubique locale comme modèle de l'écoulement. Des canaux à parois sinusoïdales à géométrie variable sont utilisés pour représenter différents types de fractures. Un premier développement analytique montre que l'ouverture hydraulique de la fracture diffère de son ouverture moyenne lorsque la rugosité des parois est élevée. La méthode des éléments finis est ensuite utilisée pour résoudre les équations de continuité et de Navier-Stokes et comparer les solutions numériques aux prédictions théoriques de la loi cubique locale sur une gamme relativement étendue de nombres de Reynolds Re. Pour de faibles Re, typiquement inférieurs à 15, la loi cubique locale décrit raisonnablement l'écoulement, surtout lorsque la rugosité et le déphasage entre les parois sont relativement faibles. Dans un deuxième temps, les écoulements chargés de particules sont étudiés. Une approche analytique est d'abord développée pour montrer comment des particules distribuées dans un écoulement stationnaire et laminaire à travers une fracture peuvent être transportées sur de longues distances ou au contraire se déposer à l'intérieur. Plus précisément, une équation simple décrivant la trajectoire d'une particule est établie. Sur la base de cette équation, il est démontré que, quand l'inertie des particules est négligeable, leur comportement dépend directement de la géométrie de la fracture et d'un nombre adimensionnel W qui relie la vitesse de sédimentation des particules à la vitesse moyenne de l'écoulement. L'équation proposée est vérifiée en comparant ses prédictions à des simulations numériques de suivi de particules prenant en compte l'inertie des particules et résolvent complètement les équations de Navier-Stokes. Il est montré que l'équation est valide lorsque l'inertie du fluide est faible. Des diagrammes de régimes, permettant de prévoir le comportement des particules à travers la fracture sont proposés. Enfin, un appareil expérimental conçu dans le but d'effectuer une évaluation pratique du modèle analytique est présenté et les résultats préliminaires sont discutés. Les résultats expérimentaux préliminaires tendent valider le modèle analytique. De façon plus générale, les résultats obtenus à travers ce travail de thèse font progresser nos connaissances du comportement des petites particules transportées dans les écoulements de fractures. Potentiellement, ce travail devrait permettre d'améliorer notre prévision de la pollution souterraine, et peut avoir des applications dans le développement de nouvelles techniques de filtration de l'eau et de séparation des minéraux / The aim of the present thesis is to study the transport and deposition of small solid particles in fracture flows. First, single-phase fracture ow is investigated in order to assess the validity of the local cubic law for modeling ow in corrugated fractures. Channels with sinusoidal walls having different geometrical properties are considered to represent different fracture geometries. It is analytically shown that the hydraulic aperture of the fracture clearly deviates from its mean aperture when the walls roughness is relatively high. The finite element method is then used to solve the continuity and the Navier-Stokes equations and to simulate fracture ow in order to compare with the theoretical predictions of the local cubic law for Reynolds numbers Re in the range 0.067-67. The results show that for low Re, typically less than 15, the local cubic law can properly describe the fracture ow, especially when the fracture walls have small corrugation amplitudes. For Re higher than 15, the local cubic law can still be valid under the conditions that the fracture presents a low aspect ratio, small corrugation amplitude, and moderate phase lag between its walls. Second, particle-laden flows are studied. An analytical approach has been developed to show how particles sparsely distributed in steady and laminar fracture flows can be transported for long distances or conversely deposited inside the channel. More precisely, a rather simple particle trajectory equation is established. Based on this equation, it is demonstrated that when particles' inertia is negligible, their behavior is characterized by the fracture geometry and by a dimensionless number W that relates the ratio of the particles sedimentation terminal velocity to the ow mean velocity. The proposed particle trajectory equation is verified by comparing its predictions to particle tracking numerical simulations taking into account particle inertia and resolving the full Navier-Stokes equations. The equation is shown to be valid under the conditions that ow inertial effects are limited. Based on this trajectory equation, regime diagrams that can predict the behavior of particles entering closed channel flows are built. These diagrams enable to forecast if the particles entering the channel will be either deposited or transported till the channel outlet. Finally, an experimental apparatus that was designed to have a practical assessment of the analytical model is presented. Preliminary experimental results tend to verify the analytical model. Overall, the work presented in this thesis give new insights on the behavior of small particles in fracture flows, which may improve our prediction and control of underground contamination, and may have applications in the development of new water filtration and mineral separation techniques

Orientador: Prof. Dr. Francisco Eugenio Mendonça da Silveira / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Física, 2014. / Neste trabalho, discutimos o efeito de uma relaxação finita para a corrente sobre a instabilidade axialmente simetrica de uma coluna de plasma com perfil parabólico na aproximacão não indutiva. A forma funcional da densidade de momento angular específico é dada por L = ar2+br..1=2, onde a e b s~ao constantes. Isto signica que L possui um valor mnimo para uma dada coordenada radial r, limitada pelas paredes do vaso que contem o uido. Como consequ^encia, o sistema pode se tornar instavel se a frequ^encia epicclica k = [2 (2 + r 0)]1=2 for um numero imaginario, onde denota a velocidade angular do uido e a linha, uma derivada total em relac~ao a r. Uma analise de instabilidade linear e desenvolvida para os limites de grande e pequena resistividade do uido. Os resultados apresentados aqui podem contribuir para tornar a noc~ao da aproximac~ao n~ao indutiva em um conceito mais preciso. Finalmente, apresentamos uma proposta de continuidade natural deste trabalho. / In this work, we discuss the efect of a fnite relaxation for the current on the axisymmetric instability of a plasma column with parabolic profile in the inductionless approximation. The functional form of the specic angular momentum density is given by L = ar2 + br..1=2, where a and b are constants. As a consequence, the system may become unstable if the epicyclic frequency k = [2 (2 + r 0)]1=2 is an imaginary number, where denotes the angular velocity of the uid and the prime, a total derivative with respect to r. A linear stability analysis is developed for the limits of large and of small resistivity of the uid. The results presented here may contribute to turn the notion of the inductionless approximation into a more precise concept. Finally, we present a proposal of natural continuity of this work.

Rising penetration of renewable energy sources in electric power grids isboth a challenge and an opportunity to optimally utilize the potential of eitherwind or PV energy sources, to stabilize operation of future power systems.Bi-directional ows between distribution and transmission system operatorscause signicant problems with keeping the voltages in the grid within admissiblelimits. This paper contains description of Oland's island mediumandlow-voltage electric power grid, ranging from 0.4 kV to 130 kV in thepurpose of quasi-static analysis of active and reactive power ows in the system.Goal of the analysis is to optimize reactive power exchange at the pointof connection with the mainland grid. In the analyzed grid system, thereis an enormous, 190 % penetration of wind sources. Capacity of the windparks connected to dedicated buses totals to 136.1 MW, that supply up to90.5 MW of load. With industry-wise reactive power capability limits, totalcontribution of wind parks reaches almost 66 MVAr, enabling to compensatedecits and extra surpluses of the reactive power in the grid. Presentedsystem is connected to the mainland's grid through one point of connection,which is simulated as Thevenin equivalent circuit. Main objective of thethesis is to test and analyze viable solutions to minimize reactive power exchangeat the point of connection at Stavlo substation connecting Oland'sand Sweden's electric grid keeping valid all necessary contingencies enforcedby current grid codes applied in Sweden as well as thermal limits of the linesand voltage limits of the system. Furthermore, state of the art of currentreactive power compensation methodologies and most promising techniquesto eciently and eectively control reactive power ow are outlined. Droopcontrol methodologies, with focus on global and local objectives, and smartgrid solutions opportunities are being tested and modeled by the authors andare comprehensively presented in this paper. Moreover, economic costs ofcontrol methods are compared. Analysis of active power losses in the systemas well as cost of implementation of alternative solutions is presented, wheremost nancially viable solutions are outlined, giving brief outlook into futureperspectives and challenges of electric power systems. It is shown that controllabilityof reactive power support by wind turbine generators can enhanceoperation of electric power grids, by keeping the reactive power ow minimizedat the boundary between grids of distribution and transmission systemoperators. Furthermore, results indicate that extra reactive power supportby wind turbine generators can lead to diminishment of active power losses inthe system. Presented system is being modeled in the PSS/E software dedicatedfor power system engineers with use of Python programming languages.Analysis of data was done either in Python or R related environments. Thesiswas written with cooperation between KTH and E.On Energidistribution AB. / Hogre genomslagskraft av förnyelsebara energikällor i elnäteten är bådeen utmaning och möjlighet för att optimalt kunna utnyttja potentialen av vindkraft och PV källor, med avseende på att stabilisera driften av framtida elkraftsystem. Tvåvägsflöden mellan distributionoch transmissionsoperatörer orsakar betydande problem att hålla spänningen i nätet inom tillåtna gränsvärden.Denna uppsats innehåller en beskrivning av Ö lands mellanoch lågspänningsnät,på 0.4 kV till 130 kV i syftet att utföra en kvasistatisk analys av aktiva och reaktiva effektflöden i systemet. Målet med analysen är att optimera det reaktiva effektutbytet i kopplingspunkten med fastlandets nät. I det analyserade systemet, finns det en enorm potential på 190% genomslagskraft av vindkraft. Kapaciteten på vindkraftsparker kopplade till medtagna samlingsskenor i systemet uppgår till 136,1 MW, som tillgodoser upp till 90.5 MW last. Med industrimässigt begränsad reaktiv effektkapabilitet, uppgår vindkraftsparkernas bidrag till nästan 66 MVAr, vilken möjliggör kompensation för underskott och överskott av reaktiv effekt i nätet. Det presenterade systemet är kopplat till fastlandet genom en kopplingspunkt, där fastlandet är simulerat som en Thevenin ekvivalent. Huvudsakliga målet med denna uppsats är att testa och analysera gångbara lösningar för att minimera det reaktiva effektutbytet vid kopplingspunkten i Stävlö, som kopplar ihop Ö land med resterande nät i Sverige, samtidigt som alla nödvändiga villkor enligt nuvarande nätkoder i Sverige bibehålls, liksom termiska gränser för ledningarna och spanningsgränser för systemet. Ytterligare beskrivs den bästa tillgängliga tekniken som finns idag för reaktiv effektkompensation, och de mest lovande teknikerna för att effektivt och verkningsfullt kontrollera reaktiva effektflöden. Droop-kontroll-metodologier, med fokus på globala och lokala tillämpningar, och smarta nät-möjligheter testas och modelleras av författarna och presenterar djupgående i detta arbete. Dessutom jämförs ekonomiska kostnader för olika kontrollmetoder. Analyser av aktiva effektförluster i systemet samt kostnader för implementation av alternativa lösningar presenteras, där de flesta gångbara losningar behandlas, och ger en överskådlig bild av framtida perspektiv och utmaningar i elkraftsystemet. Det visas att vindturbiners kontroll av reaktiv effekt, kan förbättra driften av elnäten, genom att minimera det reaktiva effektflödesutbytet i gränsen mellan distributionoch transmissionsoperatörers nät. Ytterligare pekar resultat på att extra understöd av reaktiv effekt från vindturbiner kan leda till förminskning av aktiva förluster i systemet. Det presenterade systemet modelleras i mjukvaruprogrammet PSS/E dedikerat för elkraftsingenjörer med hjälp av Python. Analys av data gjordes antingen i Pythoneller R-relaterade miljöer. Detta arbete har gjorts tillsam-mans med KTH och E.ON Energidistribution AB.