Speaker:Dr. Oscar Argudo
(ViRVIG group of the Universitat Politècnica de Catalunya, Spain)
Virtual landscapes in media and games nowadays display large impressive terrains with richness of details. Therefore, the current challenge is not only to produce visually appealing scenes, but also to ensure they conform to some objective criteria for realism. For example, we could simulate the physical processes underlying natural phenomena, procedurally mimic distributions of measured properties, or learn from real data. In this talk, I will present a few works that followed these ideas to create a variety of landscapes: from deserts to glaciers, from alpine rocky peaks to gentle forested hills, and different degradation effects on natural scenes. Apart from the knowledge borrowed from Earth Sciences and other disciplines outside Computer Science, we will see the inspiration and key ideas in many of these works came from actual hikes!
I am currently a Maria Zambrano research fellow at the ViRVIG group of the Universitat Politècnica de Catalunya. I obtained my PhD in Computing from UPC in 2018, under the supervision of Carlos Andújar and Antonio Chica. My thesis focused on the creation of realistic natural scenarios, leveraging machine learning techniques and real data to improve procedural and example-based modeling algorithms. After that, I was hired as a postdoctoral researcher by the CNRS in the LIRIS laboratory in Lyon, working on procedural modeling of mountainous landscapes and the simulation of natural phenomena such as dunes, glaciers and ecosystems. My current research project deals with the generation of hiking paths networks and the modeling of degradation effects caused by outdoor activities. I have published in journals such as ACM Transactions on Graphics and Computer Graphics Forum, and presented in top conferences like SIGGRAPH Asia and Eurographics.
In recent years there has been a lot of research in the area of edutainment, which facilitates effective learning processes by increasing the engagement of the learners. Guided visualisations, such as audio-guided museum tours or AR-guided city tours, are one of the potential applications.Guided visualisations are a form of mental practice that traditionally involves verbal guidance that guides a user through a series of visualisations. With the technique of Augmented Reality, one can integrate additional information to guide a user or use a virtual character to embody verbal guidance, which enables an engaging user experience. In this thesis, we aim to make a first step towards guided visualisation by introducing a hand-drawn character for instruction purposes. We especially focus on animation, since character animations are used in different applications, such as computer graphics, but can be hardly generated without certain pre-knowledge. Here, we present a novel pipeline for automatically generating believable movements for hand-drawn characters. The approach consists of five steps. (1) The hand-drawn character is detected from an input image, and (2) the sub-parts of the drawn character, such as the legs and the head, are identified, respectively. (3) A bone skeleton for animation is extracted and augmented with the semantic information from the previous step. (4) Based on the augmented skeleton, we assign a super-class that the skeleton belongs to, i.e. quadruped or humanoid, and match the end-effectors of the skeleton to the end-effectors of the reference skeleton of the super-class. (5) Finally, we generate a triangular mesh from the input illustration. Once the right skeleton and the hand-drawn character are overlayed, the character is animated and can attract users in different applications.To show the feasibility of our approach, we evaluate the proposed pipeline with a set of hand-drawn characters showing several well-articulate drawings.
20 + 10
Supervisor: Eduard Gröller
Institute of Visual Computing & Human-Centered Technology
Favoritenstr. 9-11 / E193-02
Austria - Europe