Hasso-Plattner-Institut
Prof. Dr. Jürgen Döllner
  
 

Semantic Image Synthesis (GauGAN) on Mobile Devices

Synopsis: GauGAN is a technique based on Generative Adversarial Network (GAN), recently introduced by NVidia, that allows drawing semantic masks that are instantly transformed into a photo-realistic image. In this thesis, the feasibility of porting this network to the iOS CoreML framework for on-device processing should be analyzed. The main challenge of this project concerns the reduction of model memory consumption to facilitate execution on mobile hardware. For it, different approaches should be evaluated: quantization of model data, light-weight model architectures trained via transfer learning, and out-of-core strategies.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)
  • Max Reimann (max.reimann(at)hpi.de)

Painterly Rendering Using Vector Heat Method

Synopsis: The goal of this master thesis is to implement an interactive abstraction technique that combines painterly rendering with the Vector Heat Method (VHM): required to align paint strokes. Starting with the base image, a simple user interface should enable painting curves on top of it. Afterward, curve tangents are transferred and aligned to the input image using the VHM. Subsequently, points are distributed over the canvas surface to avoid overlapping and strokes are aligned according to the transported vectors. The resulting strokes then can be reparameterized to achieve different abstraction effects.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)

Video-saliency based on Frame-rate

Synopsis: An important aspect of the human visual system is the ability to quickly select relevant and salient regions. This ability enables humans to easily interpret complex dynamic scenes in real-time. For it, methods for video-saliency aim at identifying important regions for observers within a video. However, most of the existing methods ignore the frame-rate while computing saliency. In this thesis, the effect of frame-rate for video-saliency estimation should be analyzed, and a frame-rate adaptive video-saliency algorithm based on machine-learning should be developed.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)
  • Sumit Shekhar (sumit.shekhar(at)hpi.de)

Human Motion Depiction from Pose Estimation

Synopsis: Recent advances in deep-learning have resulted in fast and accurate methods for detecting human poses in images and videos. The goal of this thesis is to implement approaches for detecting and analyzing complex human motion patterns in short input videos (e.g., performing sports or dancing) and generate stylized still images based on these analyses. The focus will be on tracking motion-saliency and generating a coherent motion-path with an appealing style. Additionally, the proposed method can be implemented and evaluated in an existing mobile app for motion depiction of live photos.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)
  • Max Reimann (max.reimann(at)hpi.de)

Smart Video and Image Processing for Mobile Devices

Synopsis: Video and image processing is often a demanding task with respect to processing power and memory requirements. To provide such functionality for mobile devices, the common concept of Software-as-a-Service (SaaS) is utilized, i.e., mobile devices serve as provider for input data that is processed using cloud computing. This thesis focus on developing a client application for services based video processing on iOS or Android operating systems and the VideoToolKit (VTK): a system for real-time analysis, processing, and rendering of video and image streams that is currently under development at the Computer Graphics Systems Group. The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam. 

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)
  • Sebastian Pasewaldt (sebastian.pasewaldt(at)hpi.de)

Visual Computing based on Vulkan and MoltenVK

Synopsis: While current graphics APIs are fragmented into different ecosystems (OpenGL, OpenGL-ES, DirectX, Metal), the new Vulkan standard promises true platform compatibility. In this thesis, a system for executing visual computing assets (VCAs) on mobile devices should be developed and compared to previous approaches. The major question this thesis would address is, how to effectively (and possibly automatically) adapt Vulkan-based image processing VCAs for mobile platforms, e.g., using MoltenVK on iOS. A number of reference VCAs are provided and could be integrated into existing iOS/Android frameworks of our research partner, Digital Masterpieces GmbH.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)
  • Max Reimann (max.reimann(at)hpi.de)

Transpiler for Visual Computing Assets

Synopsis: Visual Computing Assets (VCAs) are used to store, represent, and manage techniques for image and video analysis and processing in a way that is independent of platform and rendering API. VCAs are the foundation for multiple technical demonstrators and applications developed at the Computer Graphics Systems chair. The aim of this master thesis is to implement a transpiler from XML-based VCAs to web-based rendering technologies based on glTF, X3D, or X3DOM using XSLT transformations. Additionally, new technologies for browser-based machine learning such as TensorFlow.js can be integrated. Thus, enabling the application of complex VCAs for image and video processing within web-browser.

The master thesis can be used to start as a software developer at our research partner Digital Masterpieces GmbH (www.digitalmasterpieces.com) in Potsdam or for a PhD position at the Computer Graphics Systems chair. If you are interested or have any further questions please contact the research group Computer Graphics Systems:

  • Prof. Dr. Jürgen Döllner (office-doellner(at)hpi.de)
  • Dr. Matthias Trapp (matthias.trapp(at)hpi.de)