Cette thèse s’intéresse à la synthétisation de paysages naturels, et plus particulièrement, à leur portion céleste. L’aspect du ciel est gouverné par de nombreux phénomènes atmosphériques parmi lesquels les nuages jouent un rôle prépondérant car ils sont fréquemment présents et couvrent de grandes étendues. Même sans considérer directement le ciel, la densité des nuages leur permet de modifier intensément l’illumination globale d’un paysage. Les travaux de cette thèse se concentrent donc principalement sur l’édition, la modélisation et l’animation d’étendues nuageuses aux dimensions d’un paysage. Comme la simulation thermodynamique de la formation des nuages est difficilement contrôlable et que les détails du volume simulés sont rapidement limités, nous proposons plutôt une méthode par génération procédurale. Nous érigeons un modèle léger de paysage nuageux sous forme d’une hiérarchie de fonctions. Les détails les plus fins sont obtenus par composition de bruits procéduraux et reproduisent les formes de différents genres de nuages. La présence nuageuse à grande échelle est quant à elle décrite à haut niveau et à différents instants par des cartes dessinées par l’utilisateur. Ces cartes discrètes sont transformées en primitives implicites statiques ensuite interpolées par métamorphose en prenant en compte le relief et les vents pour produire des trajectoires cohérentes. Le champ implicite obtenu par mélange des primitives interpolantes constitue le champ spatiotemporel de densité nuageuse. Des images sont finalement synthétisées par rendu du milieu participatif atmosphérique selon notre propre implémentation exécutée en parallèle sur carte graphique / This thesis focusses on the synthetization of natural landscapes, and more particularly on their celestial part. The aspect of the sky is governed by plenty of atmospheric phenomena, among which clouds play a major role for they are recurrent and widespread. Even without directly considering the sky, the density of the clouds allows them to intensely modify the global illumination of a landscape. The work of this thesis thus focuses mainly on the editing, modelling and animation of cloud areas of landscape dimensions.Because the thermodynamic simulation of cloud formation is hard to control and its maximum resolution quickly limits the details of the simulated volume, we propose instead a procedural generation method. We build a lightweight cloudscape model as a hierarchy of functions. The finest details are obtained by composing procedural noises and reproduce the specific shapes of different kinds of clouds. The large-scale cloud presence is described at a high level and at different times by maps drawn by the user. These discrete maps are transformed into implicit static primitives and then interpolated by morphing, accounting for relief and winds in order to produce coherent trajectories. The implicit field obtained by mixing the interpolating primitives represents the spatiotemporal field of cloud density. Images are finally synthesized by rendering of the atmospheric participative medium according to our own implementation, executed in parallel on a graphic card

在這個研究中，我們嘗試將節奏的要素（速度、誇張度與時間調配）參數化，以產生能控制特定風格之人物角色的動畫。角色動作風格化的生成及控制是藉由一個層級式的動畫控制系統RhyCAP (Rhythmic Character Animation Playacting system), 透過一個節奏動作控制（Rhythmic Motion Control, RMC) 的方法來實現。RMC是基於傳統動畫的原則，設計參數化的動作指令，來產生生動並具有說服力的角色動作。此外，RMC也提供了運動行為的模型來控制角色動畫的演出。藉由RhyCAP系統所提供的高階控制介面，即使是沒有經過專業傳統動畫技巧訓練的使用者，也能夠創作出戲劇性的中國舞獅動畫。 / In this research, we attempt to parameterize the rhythmic factors (tempo, exaggeration and timing) into the generation of controllable stylistic character animation. The stylized character motions are generated by a hierarchical animation control system, RhyCAP (Rhythmic Character Animation Playacting system) and realized through an RMC (Rhythmic Motion Control) scheme. The RMC scheme can generate convincible and expressive character motions from versatile action commands with the rhythmic parameters defined according to the principles of traditional animation. Besides, RMC also provide controllable behavior models to enact the characters. By using the high-level control interface of the RhyCAP system, the user is able to create a dramatic Chinese Lion Dance animation intuitively even though he may not be professionally trained with traditional animation skills.

Context. Animation of character locomotion is an important part of computer animation and games. It is a vital aspect in achieving believable behaviour and articulation for virtual characters. For games there is also often a need for supporting real-time reactive behaviour in an animation as a response to direct or indirect user interaction, which have given rise to procedural solutions to generate animation of locomotion. Objectives. In this thesis the performance aspects for procedurally generating animation of locomotion within real-time constraints is evaluated, for bipeds and quadrupeds, and for simulations of several characters. A general pose-driven feedback algorithm for physics-driven character locomotion is implemented for this purpose. Methods. The execution time of the locomotion algorithm is evaluated using an automated experiment process, in which real-time gait simulations of incrementing character population count are instantiated and measured, for the bipedal and quadrupedal gaits. The simulations are measured for both serial and parallel executions of the locomotion algorithm. Results. Simulations of up to and including 100 characters are performance measured providing an overview of the slowdown rate when increasing the character count in the simulations, as well as the performance relations between bipeds and quadrupeds. Conclusions. The experiment concludes that the evaluated algorithm on its own exhibits a relatively small performance impact that scales almost linearly for the evaluated population sizes. Due to the relatively low performance impacts it is thus also concluded that for future experiments a broader measurement of the locomotion algorithm that includes and compares different physics solvers is of interest.

Various research has shown the potential and robustness of deep learning-based approaches to synthesise novel motions of 3D characters in virtual environments, such as video games and films. The models are trained with the motion data that is bound to the respective character skeleton (rig). It inflicts a limitation on the scalability and the applicability of the models since they can only learn motions from one particular rig (domain) and produce motions in that domain only. Transfer learning techniques can be used to overcome this issue and allow the models to better adapt to other domains with limited data. This work presents a study of three transfer learning techniques for the proposed Objective-driven motion generation model (OMG), which is a model for procedurally generating animations conditioned on positional and rotational objectives. Three transfer learning approaches for achieving rig-agnostic encoding (RAE) are proposed and experimented with: Feature encoding (FE), Feature clustering (FC) and Feature selection (FS), to improve the learning of the model on new domains with limited data. All three approaches demonstrate significant improvement in both the performance and the visual quality of the generated animations, when compared to the vanilla performance. The empirical results indicate that the FE and the FC approaches yield better transferring quality than the FS approach. It is inconclusive which of them performs better, but the FE approach is more computationally efficient, which makes it the more favourable choice for real-time applications. / Många studier har visat potentialen och robustheten av djupinlärningbaserade modeller för syntetisering av nya rörelse för 3D karaktärer i virtuell miljö, som datorspel och filmer. Modellerna är tränade med rörelse data som är bunden till de respektive karaktärskeletten (rig). Det begränsar skalbarheten och tillämpningsmöjligheten av modellerna, eftersom de bara kan lära sig av data från en specifik rig (domän) och därmed bara kan generera animationer i den domänen. Kunskapsöverföringsteknik (transfer learning techniques) kan användas för att överkomma denna begränsning och underlättar anpassningen av modeller på nya domäner med begränsade data. I denna avhandling presenteras en studie av tre kunskapsöverföringsmetoder för den föreslagna måldriven animationgenereringsnätverk (OMG), som är ett neural nätverk-baserad modell för att procedurellt generera animationer baserade på positionsmål och rotationsmål. Tre metoder för att uppnå rig-agnostisk kodning är presenterade och experimenterade: Feature encoding (FE), Feature clustering (FC) and Feature selection (FS), för att förbättra modellens lärande på nya domäner med begränsade data. All tre metoderna visar signifikant förbättring på både prestandan och den visuella kvaliteten av de skapade animationerna, i jämförelse med den vanilla prestandan. De empiriska resultaten indikerar att både FE och FC metoderna ger bättre överföringskvalitet än FS metoden. Det går inte att avgöra vilken av de presterar bättre, men FE metoden är mer beräkningseffektiv, vilket är fördelaktigt för real-time applikationer.